Primary versus secondary contributions to particle numberconcentrations in the European boundary layer
C. L. Reddington1, K. S. Carslaw1, D. V. Spracklen1, M. G. Frontoso2, L. Collins1, J. Merikanto1,3, A. Minikin 4,T. Hamburger4, H. Coe5, M. Kulmala 3, P. Aalto3, H. Flentje6, C. Plass-Dulmer6, W. Birmili 7, A. Wiedensohler7,B. Wehner7, T. Tuch7, A. Sonntag7, C. D. O’Dowd8, S. G. Jennings8, R. Dupuy8, U. Baltensperger9, E. Weingartner9,H.-C. Hansson10, P. Tunved10, P. Laj11, K. Sellegri12, J. Boulon12, J.-P. Putaud13, C. Gruening13, E. Swietlicki14,P. Roldin14, J. S. Henzing15, M. Moerman15, N. Mihalopoulos16, G. Kouvarakis16, V. Zdımal17, N. Zıkova17,A. Marinoni 18, P. Bonasoni18, and R. Duchi18
1Institute for Climate and Atmospheric Science, School of Earth and Environment, University of Leeds, Leeds, UK2C2SM – ETH Zurich, Zurich, Switzerland3Division of Atmospheric Sciences, Department of Physics, University of Helsinki, Helsinki, Finland4Deutsches Zentrum fur Luft- und Raumfahrt (DLR), Institut fur Physik der Atmosphare, Oberpfaffenhofen, Germany5School of Earth, Atmospheric and Environmental Sciences, University of Manchester, Manchester, UK6Deutscher Wetterdienst, Meteorologisches Observatorium Hohenpeißenberg, Germany7Leibniz Institute for Tropospheric Research, Leipzig, Germany8School of Physics & Centre for Climate and Air Pollution Studies, Ryan Institute, National University of Ireland Galway,University Road, Galway, Ireland9Paul Scherrer Institut, Laboratory of Atmospheric Chemistry, 5232 Villigen, Switzerland10Institute for Applied Environmental Research, Stockholm University, Stockholm, Sweden11UJF-Grenoble 1/CNRS, LGGE UMR5183, Grenoble 38041, France12Laboratoire de Meteorologie Physique, CNRS, Universite Blaise Pascal, Aubiere cedex, France13European Commission, Joint Research Centre, Institute of Environment and Sustainability, Ispra, Italy14Division of Nuclear Physics, Lund University, P.O. Box 118, 22100 Lund, Sweden15Netherlands Organisation for Applied Scientific Research TNO, Utrecht, The Netherlands16Environmental Chemical Processes Laboratory, Department of Chemistry, University of Crete, Heraklion, Crete, Greece17Institute of Chemical Process Fundamentals of the AS CR, Prague, Czech Republic18CNR-Institute for Atmospheric Sciences and Climate, Bologna, Italy
Received: 31 May 2011 – Published in Atmos. Chem. Phys. Discuss.: 28 June 2011Revised: 11 November 2011 – Accepted: 16 November 2011 – Published: 5 December 2011
Abstract. It is important to understand the relative contri-bution of primary and secondary particles to regional andglobal aerosol so that models can attribute aerosol radiativeforcing to different sources. In large-scale models, there isconsiderable uncertainty associated with treatments of parti-cle formation (nucleation) in the boundary layer (BL) andin the size distribution of emitted primary particles, lead-ing to uncertainties in predicted cloud condensation nuclei
(CCN) concentrations. Here we quantify how primary par-ticle emissions and secondary particle formation influencesize-resolved particle number concentrations in the BL usinga global aerosol microphysics model and aircraft and groundsite observations made during the May 2008 campaign of theEuropean Integrated Project on Aerosol Cloud Climate AirQuality Interactions (EUCAARI). We tested four differentparameterisations for BL nucleation and two assumptions forthe emission size distribution of anthropogenic and wildfirecarbonaceous particles. When we emit carbonaceous par-ticles at small sizes (as recommended by the Aerosol Inter-comparison project, AEROCOM), the spatial distributions of
Published by Copernicus Publications on behalf of the European Geosciences Union.
12008 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
campaign-mean number concentrations of particles with di-ameter>50 nm (N50) and>100 nm (N100) were well cap-tured by the model (R2≥0.8) and the normalised mean bias(NMB) was also small (−18 % forN50 and−1 % for N100).Emission of carbonaceous particles at larger sizes, which weconsider to be more realistic for low spatial resolution globalmodels, results in equally good correlation but larger bias(R2
≥0.8, NMB =−52 % and−29 %), which could be partlybut not entirely compensated by BL nucleation. Within theuncertainty of the observations and accounting for the uncer-tainty in the size of emitted primary particles, BL nucleationmakes a statistically significant contribution to CCN-sizedparticles at less than a quarter of the ground sites. Our re-sults show that a major source of uncertainty in CCN-sizedparticles in polluted European air is the emitted size of pri-mary carbonaceous particles. New information is requirednot just from direct observations, but also to determine the“effective emission size” and composition of primary parti-cles appropriate for different resolution models.
1 Introduction
Atmospheric aerosol particles are generally classified as ei-ther primary or secondary depending on their source or ori-gin. Increases in the number concentrations of primary andsecondary aerosol from anthropogenic sources have beenshown to increase the number concentrations of cloud con-densation nuclei (CCN) and cloud drops (e.g. Ramanathanet al., 2001), potentially modifying the properties of clouds(e.g. Lohmann and Feichter, 2005). However, there are largeuncertainties associated with the primary emission fluxes andsecondary formation rates of atmospheric aerosol, leading touncertainties in predicted global CCN concentrations (Pierceand Adams, 2009; Merikanto et al., 2009) and ultimatelycloud radiative forcing.
Primary particles are emitted directly into the atmospherefrom natural sources such as volcanoes, forest fires, seaspray, and windborne dust, and anthropogenic sources suchas fossil fuel burning in combustion engines and powerplants. Primary particle emissions are estimated to contributeabout 55 % of global CCN number concentrations at 0.2 %supersaturation (CCN (0.2 %)) in the boundary layer (BL),and up to 70 % in polluted continental regions (Merikantoet al., 2009). However, Merikanto et al. (2009) also showedthat the estimated contribution of primary particles to CCNis uncertain due to uncertainties in the size distribution ofthe emitted particles. Aerosol modelling studies often usedifferent parameterisations for the prescribed emission sizedistribution (e.g.Textor et al., 2006), leading to significantdifferences in modelled primary particle number and thus es-timated CCN number concentrations (Spracklen et al., 2010).Spracklen et al. (2011) demonstrate that primary carbona-ceous particles make an important contribution to the aerosol
indirect effect, but estimates vary by a factor of∼3 depend-ing on the prescribed emission size distribution.
Secondary aerosol particles are formed in the atmospherethrough homogeneous nucleation (gas-to-particle conver-sion) of both natural and anthropogenic gaseous precursors.Once formed, a fraction of nucleated particles undergo sub-sequent growth through condensation of gas-phase speciesand self-coagulation, and have the potential to reach parti-cle sizes relevant for CCN and cloud drop formation (Ker-minen et al., 2005). Secondary aerosol formation has beenobserved to occur globally over many different regions bothwithin the BL and in the upper free troposphere (FT) (seeKulmala et al., 2004, and references therein). Observations(Lihavainen et al., 2003; Laaksonen et al., 2005) and mod-elling studies (Spracklen et al., 2008; Merikanto et al., 2009;Wang and Penner, 2009; Yu and Luo, 2009) have shownthat secondary particles make important contributions to re-gional and global CCN concentrations. Globally, 45 % ofCCN (0.2 %) in the BL are estimated to derive from nucle-ation (Merikanto et al., 2009), although again this number isuncertain (range 31–49 %) due to uncertainties in nucleationrates and the properties of the primary particles. The uncer-tainties estimated in Merikanto et al. (2009) may be too lowsince they did not take into account the multiple plausiblenucleation mechanisms (e.g. Spracklen et al., 2010; Metzgeret al., 2010; Paasonen et al., 2010; Yu et al., 2010).
The process of binary homogeneous nucleation (BHN) ofwater and sulphuric acid (Kulmala and Laaksonen, 1990;Kulmala et al., 1998; Vehkamaki et al., 2002), with its strongtemperature dependence, is able to reproduce high particleconcentrations observed in the cold free and upper tropo-sphere (Adams and Seinfeld, 2002; Spracklen et al., 2005a).But in the warmer lower troposphere, production rates arelow (Lucas and Akimoto, 2006). Additional mechanismshave been suggested to explain observed particle formationsuch as ternary nucleation of water, sulphuric acid and am-monia (Kulmala et al., 2000; Anttila et al., 2005; Merikantoet al., 2007); multi-component nucleation with the partici-pation of organics instead of ammonia (e.g. Metzger et al.,2010); and ion-induced nucleation (Laakso et al., 2002;Modgil et al., 2005). However, with the exception of organ-ics, their contribution to secondary particle concentrationsin the continental BL is thought to be fairly limited (Anttilaet al., 2005; Laakso et al., 2007; Kulmala et al., 2007; Boyet al., 2008; Elleman and Covert, 2009).
Observations of BL nucleation events at various Europeansurface measurement sites have revealed a strong correlationbetween the measured particle formation rate and the gas-phase concentration of sulphuric acid to the power of oneor two (e.g. Sihto et al., 2006; Riipinen et al., 2007; Paaso-nen et al., 2009, 2010). By measuring newly formed par-ticles (∼1.5 nm in diameter) in the laboratory, Sipila et al.(2010) have recently confirmed the linear and squared re-lationships between nucleation rate and sulphuric acid con-centration that are observed in the atmosphere. These
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12009
observations have been used to develop empirical nucleationrates, where the formation rate of sub-3 nm molecular clus-ters (Jnuc) is related to the gas-phase sulphuric acid concen-tration ([H2SO4]) with either a linear i.e.Jnuc= A[H2SO4],or a squared i.e.Jnuc= K[H2SO4]
2 dependence (e.g. Weberet al., 1996; Kulmala et al., 2006; Sihto et al., 2006; Riipinenet al., 2007).
To describe the observed linear dependence, Kulmala et al.(2006) propose an activation mechanism, where neutral orion clusters containing one sulphuric acid molecule are acti-vated for further growth. McMurry and Friedlander (1979)explain the squared dependence by proposing a kinetic nu-cleation mechanism. The values of the nucleation rate coef-ficientsA andK; derived from surface observations of par-ticle formation events, vary spatially and temporally in theEuropean BL (e.g. Sihto et al., 2006; Riipinen et al., 2007).Riipinen et al. (2007) find that rate coefficients differ by∼4–5 orders of magnitude between different European measure-ment sites:A = 3.3×10−8
−3.5×10−4 s−1 (for the activa-tion mechanism) andK = 2.4×10−15
−1.3×10−10 cm3 s−1
(for the kinetic mechanism). A model analysis of global par-ticle number concentrations using such empirical relations(Spracklen et al., 2010) shows reasonable agreement with ob-servations at many worldwide sites, albeit with unexplainedbiases at some sites.
Other condensable vapours such as organic compoundsmay also influence the nucleation rate (e.g. Metzger et al.,2010; Paasonen et al., 2010; Kerminen et al., 2010). Paa-sonen et al. (2010) present several nucleation mechanismsthat are analogous to the kinetic- and activation-type nu-cleation theories, but consider the participation of low-volatility organic compounds in the cluster formation pro-cess both in addition to sulphuric acid and as the exclusivenucleating vapour. When evaluated against measurementsfrom European ground sites, Paasonen et al. (2010) find themost promising mechanism involves homogeneous (kinetic-type) nucleation of sulphuric acid both homomolecularly andheteromolecularly with the low-volatility organic vapours(Jnuc= k1[H2SO4]
2+k2[H2SO4][organic]). In a laboratory
study, Metzger et al. (2010) find measured particle forma-tion rates are proportional to the product concentrations ofH2SO4 and a molecule of an organic condensable species(Jnuc= k[H2SO4][organic]). Parameterising this process ina global aerosol model showed improved agreement with am-bient observations compared to control runs (Metzger et al.,2010).
In this study, we use the same aerosol microphysical modelas Spracklen et al. (2010) and extensive observations of Eu-ropean aerosol to perform a more in depth study of primaryand secondary aerosol focussing on the European BL. Weaim to better understand the absolute and relative contri-butions of primary and secondary particles to particle con-centrations over Europe, and how the contributions varyacross the particle size distribution (nucleation, Aitken andaccumulation mode sizes). We test different parameterisa-
tions for BL nucleation (including the recently proposed or-ganic/sulphuric acid nucleation mechanisms in addition tothe widely used activation and kinetic nucleation mecha-nisms), and different assumptions about the sizes and numberconcentrations of primary particle emissions that are typicalfor global aerosol and climate models. To evaluate the model,we use surface-based and airborne measurements of totalparticle number concentrations and size distribution from theIntensive Observation Period (conducted in May 2008) of theEuropean Integrated Project on Aerosol Cloud Climate AirQuality Interactions (EUCAARI; Kulmala et al., 2009). Thisstudy is a demanding test for a relatively low spatial resolu-tion global model against intensive observations in a partic-ular meteorological setting – in this case a highly pollutedanti-cyclonic period with a transition to a more dynamic sit-uation.
2 The EUCAARI intensive observation period
2.1 Aircraft and surface-based observations
A key phase of the EUCAARI Intensive Observation Pe-riod (IOP) was the Long Range Experiment (LONGREX),during which in-situ and remote sensing aerosol measure-ments were performed by the DLR Falcon 20 research air-craft, operating between 6 and 24 May 2008. Particle numberconcentrations with diameter (Dp) >4 nm (N4) and>10 nm(N10) were measured onboard the Falcon aircraft using twocondensation particle counters (CPC, TSI models 3760Aand 3010). The number concentration of non-volatile par-ticles (Dp>14 nm) was measured using an additional CPCwith a thermodenuder inlet set to a temperature of 250◦C(Burtscher et al., 2001). The total particle and non-volatileresidual size distributions were measured in the dry sizerangeDp∼0.16–6 µm using a Passive Cavity Aerosol Spec-trometer Probe-100X (PCASP; e.g. Liu et al., 1992) andGrimm Optical Particle Counter (OPC), respectively. CPCand PCASP measurements were used to calculate particlenumber concentrations in three size ranges 4–10 nm, 10–160 nm and 160–1040 nm that are roughly representative ofthe nucleation, Aitken and accumulation mode size classes,respectively. Measurements from 15 flights have been usedin this study; the tracks of these flights are shown in Fig. 1(flight sections where the altitude of the aircraft was at orbelow 2000 m a.s.l. are shown in bold).
The IOP also included spatially extensive surface-basedmeasurements from the European Supersites for Atmo-spheric Aerosol Research (EUSAAR; www.eusaar.net) andfrom the German Ultrafine Aerosol Network (GUAN; Bir-mili et al., 2009). The 15 ground sites selected for this study(see Table 1 and Fig. 1) are spread across Europe and includecoastal, boreal forest, mountain, and rural environments, andsample a range of air masses from polluted to remote conti-nental and marine. A brief description of each site is given
12010 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Table 1. Summary of surface observation sites used in this study. Site descriptions are based on the information provided by EUSAAR(www.eusaar.net) and on the site-categorisation of Henne et al. (2010).
Ground site Acronym Altitude Aerosol Description(m a.s.l.) instrument
Aspvreten, ASP 30 DMPS Boreal forest environment. Representative ofSweden regional background in Mid-Sweden.
Cabauw, CBW 60 SMPS Rural polluted environment. Air masses rangethe Netherlands from clean maritime to continental polluted.
Finokalia, FKL 250 SMPS Coastal environment. Air masses are representativeGreece of synoptic scale atmospheric composition.
Hohenpeissen- HPB 980 SMPS Rural environment. Representative of continentalberg, Germany background air masses.
Hyytiala, HTL 181 DMPS Remote, boreal forest environment. Air masses areFinland dominated by European pollution but at times very
clean Arctic air.
Jungfraujoch, JFJ 3580 SMPS Remote, high altitude site. Representative ofSwitzerland background air masses above a continental area.
JRC-Ispra, JRC 209 DMPS Semi-rural polluted environment. RepresentativeItaly of polluted continental background air masses.
K-puszta, KPO 125 DMPS Rural environment. Representative of regionalHungary background in Central-Eastern Europe.
Kosetice, KTC 534 SMPS Rural environment. Representative of continentalCzech Republic background air masses.
Melpitz, MPZ 87 DMPS Rural environment. Representative of ruralGermany polluted continental air masses.
Monte Cimone, MTC 2165 DMPS High altitude site. Representative of free troposphereItaly for South Europe/North Mediterranean area.
Puy de Dome, PDD 1465 SMPS High altitude site. Representative of regionalFrance (polluted) atmospheric background air masses.
Schauinsland, SLD 1205 SMPS Mountain ridge site (night-time site is usually aboveGermany BL, daytime site is mostly within BL), rural
environment. Representative of continentalbackground air masses.
Vavihill, VHL 172 DMPS Rural environment. Representative of continentalSweden background air masses.
in Table 1. More detailed information on the location of eachsite and the particle number concentrations observed can befound in the overview article of Asmi et al. (2011).
Diurnal variation of BL height means that the high-altitudemountain sites may not be located in the BL at all times.Therefore, without detailed screening, measurements at thesesites will not be fully representative of aerosol in the Euro-pean BL. Although this study focuses on the BL, it is im-portant to include these measurements to obtain a detailedoverview of aerosol number concentrations over Europe dur-ing the IOP. Variations in BL height are simulated in the
model used here, but have not been evaluated specifically atthe ground sites in this study. In addition to variations inBL height, the particle number concentrations measured atmountain sites may also be influenced by thermal winds orforced convection (Weingartner et al., 1999; Venzac et al.,2009), resulting in diurnal cycles in aerosol, which a rela-tively coarse resolution global model, like the one used here,is unable to capture.
Fig. 1. Map of flight tracks performed by the DLR Falcon 20 re-search aircraft during the EUCAARI-LONGREX field campaignin May 2008. Sections of the DLR Falcon flight tracks that areat or below 2 km are shown in bold. Orange dots mark the loca-tions of the European Supersites for Atmospheric Aerosol Research(EUSAAR) and the German Ultrafine Aerosol Network (GUAN)ground sites with aerosol number size distribution measurementsfor May 2008 (site acronyms are listed in Table 1).
Measurements of the aerosol particle number size distribu-tion were made using either a Scanning (SMPS) or Differen-tial Mobility Particle Spectrometer (DMPS) (e.g. Wang andFlagan, 1990) with minimum detection limits in the diameterrange 3–13 nm. Most instruments were operated accordingto the EUSAAR recommendations for mobility spectrome-ters (Wiedensohler et al., 2010), which ensure a maximumcomparability of the data collected at different measurementsites. A particular requirement is particle sizing at low rel-ative humidities (<40 %). A Europe-wide intercomparisonof instruments by the same authors showed that under de-fined laboratory conditions, the number size distributions ofsuch instruments were equivalent within±10 % for the di-ameter range 20–200 nm. Below 20 nm the uncertainty in-creases considerably. To reduce the uncertainty in the ob-servations, we restrict our analysis to the measured numbersize distribution above 15 nm. Total particle number concen-trations withDp>15 nm were calculated from the observedsize distribution.
To compare the model to the aircraft and surface obser-vations, we linearly interpolate the simulated data along theflight path of the aircraft and to the horizontal location ofthe ground site (using the model level corresponding to thealtitude of the site). The same minimum cut-off size of theinstruments (see above) is also applied to the model. Prior toanalysis, simulated data corresponding to periods of missingmeasurement data were removed. All particle number con-centrations are reported at ambient temperature and pressure.
To compare model and observations we use the normalisedmean bias (NMB) statistic:
NMB(%)=
∑ni=1(Si −Oi)∑n
i=1Oi
×100
where Si and Oi are the simulated and observed particlenumber concentrations, respectively. For comparison withthe aircraft and surface observations over the IOP, the NMB,correlation coefficient (R2), and slope of the linear regres-sion (m) are calculated between the campaign-mean mod-elled and observed number concentration from each flight oreach ground site,i. In addition, we calculate the NMB andR2 between the hourly-mean observed and simulated num-ber concentrations at each ground site (wherei representsthe hour), denoted by NMBhourly andR2
hourly.
2.2 Synoptic conditions
During the first half of the IOP (∼1–15 May 2008, hereafterPeriod A) the meteorological conditions over Central Europewere dominated by a relatively static anticyclonic blockingevent. Relatively dry and stable conditions led to an accu-mulation of European aerosol pollution inside the BL withinthe centre of the high pressure system (Hamburger et al.,2011). High particle number concentrations were observedat the surface during Period A (see Sect. 4.4). The synop-tic conditions during the second half of the IOP (∼16–31May 2008, hereafter Period B) were dominated by passageof a number of frontal systems over Central Europe. Thesesystems resulted in an increase in precipitation and a reduc-tion in both the condensation sink and particle number con-centrations, observed at the majority of the Central Europeanground sites. Hamburger et al. (2011) provide a more de-tailed description of the synoptic and pollution situation overEurope during May 2008.
3 Model description
The Global Model of Aerosol Processes (GLOMAP)(Spracklen et al., 2005a,b) simulates the evolution of size andcomposition resolved aerosols, including their interactionwith trace gases and clouds. The host model for GLOMAPis the TOMCAT global 3-D off-line Eulerian chemical trans-port model (CTM) (Chipperfield, 2006). Large scale atmo-spheric transport and meteorology in TOMCAT is specifiedfrom European Centre for Medium-Range Weather Forecasts(ECMWF) analyses, updated every 6 h. Turbulent mixing inthe BL and BL height are calculated using the parameteri-sation of Holtslag and Boville (1993). All the results havea horizontal resolution of 2.8◦×2.8◦ and 31 vertical levelsbetween the surface and 10 hPa. The vertical resolution inthe BL ranges from∼60 m near the surface to∼400 m at∼2 km a.s.l.
Here, we use GLOMAP-bin in which the aerosol size dis-tribution is specified in terms of a two-moment sectional
12012 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
(bin) scheme with 20 bins spanning 3 nm to 10 µm dry di-ameter. The aerosol particles undergo microphysical pro-cesses (coagulation, condensational growth and in-cloud pro-cessing) that alter the aerosol number size distribution in themodel. The processes of dry deposition and in-cloud/below-cloud aerosol scavenging and deposition act to remove theaerosol particles. In the following sections, we describethe features of the model that are relevant to this study.For a more detailed model description see Spracklen et al.(2005a,b).
3.1 Gas-phase emissions and chemistry
SO2 emissions are from industrial, power-plant, domestic,shipping, road transport, and off-road sources following Co-fala et al. (2005) and from volcanic sources from Andres andKasgnoc (1998). Oceanic emissions of DMS are calculatedusing the database of Kettle and Andreae (2000) and the sea-to-air transfer velocity according to Nightingale et al. (2000).Gas-phase sulphuric acid is calculated using a simplified sul-phur cycle scheme based on 7 reactions involving SO2, DMS,MSA and other minor species (Spracklen et al., 2005a). Con-centrations of oxidants OH, O3 and NO3 and HO2 are speci-fied using 6-hourly monthly-mean 3-D gridded concentrationfields from a TOMCAT simulation with detailed troposphericchemistry (Arnold et al., 2005). The oxidants are read in at 6-h intervals and linearly interpolated onto the model timestep.Emissions of biogenic terpenes are specified by the GEIA in-ventory (Benkovitz et al., 1996) and are based on Guentheret al. (1995).
3.2 Primary particles
We include emissions of primary carbonaceous aerosol fromanthropogenic sources (fossil fuel (FF) and biofuel (BF)burning) following Bond et al. (2004); and biomass burn-ing following van der Werf et al. (2003). There are somedifficulties in defining the type of carbonaceous species inan aerosol model since the definition is based upon the mea-surement technique e.g. light absorption. The carbonaceousaerosol fraction is defined by Bond et al. (2004) to consist of:black carbon (BC; the mass of combustion-generated, sp2-bonded carbon that absorbs the same amount of light as theemitted particles) and organic carbon (OC), simply the massof carbon that is not BC. It is important to note that Bondet al. (2004) treat all elemental carbon measurements as BC.Henceforth, we refer to the carbonaceous combustion aerosolas BC+OC.
Emission inventories of BC+OC particles used in largescale models are typically mass based (e.g. Cooke et al.,1999; Bond et al., 2004). To estimate the emitted parti-cle number concentration, size resolving models typicallyassume that particles are emitted with a fixed log-normalsize distribution with a specified peak number concentration(number median diameter,D) and distribution width (stan-
dard deviation,σ ). The assumption of an initial size distribu-tion for primary particles in global models accounts for boththe size of particles at emission and sub-grid scale aerosolprocesses and dynamics that influence the size and numberconcentrations of particles shortly after emission (Jacobsonand Seinfeld, 2004; Pierce et al., 2009). The assumed log-normal size distribution is also necessary to account for thelarge variability in the emission size of primary carbonaceousparticles from different sources (e.g. Bond et al., 2006, Ta-ble 3). In GLOMAP, the primary particles are “emitted”assuming an initial size distribution and then the size andnumber of particles are allowed to evolve during atmospherictransport.
The choice of the effective emission size distribution inmodels is crucial since it not only governs the emitted par-ticle number concentrations, but also affects microphysicalaerosol processes that are size-dependent. However, thereis a large range in values assumed by modellers forD
(mass median diameters for BC and OC range from∼25to ∼850 nm, Textor et al., 2006). This range has impor-tant implications for the simulated number concentrationsof primary BC+OC particles (e.g. Spracklen et al., 2010),and predicted climate-relevant quantities such as CCN andaerosol optical depth, therefore increasing the uncertainty inestimates of aerosol radiative forcing (Bauer et al., 2010).As far as the authors are aware, recommended values ofD
andσ specifically for large-scale models have only been pro-vided by Dentener et al. (2006) as part of the Aerosol Inter-comparison project (AEROCOM; http://nansen.ipsl.jussieu.fr/AEROCOM/). Grid-level and size-resolved particulateemission factors for traffic sources have been provided byZhang et al. (2005), but the grid scale used (∼300 m) is farsmaller than the grid box size of most large-scale models.
One aim of our study is to test the sensitivity of the mod-elled aerosol over Europe to the size distribution of the emit-ted anthropogenic and wildfire BC+OC. Keeping the emis-sion mass fixed, we test two sets of parameters for the log-normal size distribution that are widely used in global aerosolmodelling (shown in Fig. 2): those recommended by AERO-COM (fossil fuel emissions:DFF = 30 nm,σFF = 1.8; wild-fire and biofuel emissions:DBF = 80 nm,σBF = 1.8) (Den-tener et al., 2006); and those used by Stier et al. (2005)(DFF= 60 nm,σFF= 1.59;DBF = 150 nm,σBF = 1.59). Thefactor ∼2 difference in the recommended values forD im-plies very different BC+OC number concentrations (for fixedmass); AEROCOM requiring emitted number concentrationsto be a factor∼8 higher than Stier et al. (2005) for fossil fu-els.
The emission size distribution used by Stier et al. (2005)has been adapted from AEROCOM recommendations to fitthe standard deviation of the size modes in their model. Asa result, the spread of the primary distribution in Stier et al.(2005) (σ= 1.59) is considerably smaller than the spread ofthe AEROCOM-recommended distribution (σ= 1.8). Re-ducing the spread of the assumed emission size distribution
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12013
DFFDBFDBF
dN
/dlo
gD
10
Dry diameter (nm)
AEROCOM(Dentener et al., 2006)
Stier et al. (2005)DFF
Fig. 2. Normalised log-normal size distributions used in GLOMAPto calculate carbonaceous particle number concentrations from fos-sil fuel (FF; solid line) and biofuel (BF; dashed line) emissions.Shown are two sets of log-normal size distribution parameters(number median diameter (D) and standard deviation) from AERO-COM (Dentener et al., 2006) and Stier et al. (2005) that are widelyused in global aerosol modelling.
from σ = 1.8 toσ = 1.59, increases the emitted number con-centration by a factor of∼ 1.8, if D were constant. It is im-portant to note that in GLOMAP-bin, we are free to specifyany shape distribution within the resolution offered by the 20size bins, but use the two values ofσ as specified above. Thedifference in the parameters assumed by Stier et al. (2005)and Dentener et al. (2006) corresponds to an overall factor∼ 4.4 difference in the emitted number concentrations of fos-sil fuel BC+OC particles.
The mean size of primary FF emissions recommended byDentener et al. (2006) (DFF = 30 nm) is based on kerbsideand urban background measurements in several Europeancities (Putaud et al., 2004; Van Dingenen et al., 2004), wheretraffic-related number size distributions were dominated bya mode atDp = 20–30 nm. Although the emitted mass isgenerally conserved during transport and dispersion over theGLOMAP grid box (∼200 km at European latitudes), thenumber size distribution of primary particles shortly afteremission can be altered significantly by (sub-grid scale) at-mospheric dynamic processes such as dilution, condensa-tional growth, heterogeneous and self-coagulation, evapo-ration, and nucleation (e.g. Kittelson, 1998; Wehner et al.,2002; Zhu et al., 2002; Zhang and Wexler, 2004; Zhang etal., 2004; Roldin et al., 2010). Explicit modelling of thesesubgrid-scale processes would be too computationally ex-pensive for a global CTM, which is why an assumption ofan initial size distribution is necessary for primary BC+OCparticles.
The sub-grid evolution of the primary carbonaceous par-ticle size distribution makes it difficult to constrain the ini-tial size and particle number concentration appropriate foremission of BC+OC particles in a large model grid boxfrom measurements obtained relatively close to the emissionsource. An 85 % increase in particle diameter from the streetcanyon to the urban background was observed by Wehner
et al. (2002), which suggests the mean size of primary par-ticles is likely to increase over the model grid box from themode diameter measured at the kerbside (DFF = 30 nm). Inaddition, the statistical analysis of multiple-site observationsby Costabile et al. (2009) revealed that the coupling of ur-ban and rural number size distributions is very strong in themass-dominating accumulation mode range, but only modestin the Aitken mode range.
The size distribution of primary BC+OC particle emis-sions averaged over the model grid box is likely to be morerepresentative of the evolved size distribution of primary car-bonaceous aerosol measured at rural background sites. Itis important to note, however, that the grid-box mean sizedistribution will not necessarily correspond to the measuredparticle size at point locations. Measurement sites will belocated at varying distances from aerosol emission sourceswhich means the average processing time of the primaryaerosol will also vary between sites, thereby influencing thephysical properties of the particles measured. At sites wherethe observed particles are generally less processed than atother sites, assuming a smaller initial size for BC+OC parti-cles may agree better with the observations and vice versa.
We encounter further uncertainty associated with the as-sumed size distribution for primary BC+OC emissions whenwe consider the composition of the emitted particles. Manyaerosol models assume a homogeneous size distribution foremitted primary BC and OC (e.g.Stier et al., 2005; Textoret al., 2006, Table 4 and references therein), but the mediansizes of the BC and OC components are likely to differ in re-ality. The traffic-related ultrafine mode in the rangeDp∼3–30 nm is thought to be mostly made up of semi-volatile or-ganic compounds formed during dilution and rapid coolingof exhaust emission gases (Kittelson, 1998; Baltenspergeret al., 2002). These particles may also contain carbon com-pounds (Kittelson, 1998), and can be broadly classed as pri-mary organic matter (or OC) in the model. On the other hand,the peak emission diameter of the primary soot (BC) compo-nent is more likely to be around∼50 nm or larger as observedby Baltensperger et al. (2002). A second mode, with a maxi-mum in the rangeDp∼40–120 nm, is observed in on-road,kerbside, and urban background number size distributions(e.g. Kittelson et al., 2000, 2006; Geller et al., 2005; Casatiet al., 2007; Wehner et al., 2009; Weimer et al., 2009) and isassociated with direct emissions of soot (BC) particles fromdiesel and gasoline vehicles (e.g. Harris and Maricq, 2001).
Indications of the number concentration and size of pri-mary particles from combustion sources (such as soot) canalso be gathered from the non-volatile residues of the parti-cle number size distribution (e.g. Wehner et al., 2004; Roseet al., 2006; Engler et al., 2007; Birmili et al., 2010). Inthe urban atmosphere of Augsburg, Birmili et al. (2010)identified a clear non-volatile particle mode having a ge-ometric mean diameter between 60 and 90 nm in numberrepresentation and around 200 nm in volume representation,which Rose et al. (2006) suggest is representative of direct
12014 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
vehicular soot emissions. Engler et al. (2007) observe 2–3 non-volatile modes in the rural background particle num-ber size distribution, but the mode they associate with pri-mary emissions from combustion has typical modal diam-eters between∼70 and 90 nm. These observations suggestthat although atmospheric processes (such as dilution withthe background aerosol and/or aerosol dynamical processes)remove the number distribution fingerprint of urban primaryemissions, the mode diameter of combustion generated sootparticles remains roughly between∼50 and∼90 nm movingfrom the urban environment to the rural background.
The measurement studies discussed above suggest thatthe peak emission diameter of the BC component of traf-fic emissions is larger than the log-normal mode diame-ter recommended by Dentener et al. (2006) for fossil fuelsources, and that the mode diameter used by Stier et al.(2005) (DFF = 60 nm) may be more suitable. Yu and Luo(2009) come to a similar conclusion about both the FF andBF emission sizes recommended by AEROCOM and assumevalues ofDFF = 60 nm andDBF = 150 nm (σ= 1.8). How-ever, assuming a larger emission size that is consistent withmeasurements of primary BC/non-volatile particles may ne-glect possible contributions to the total size distribution fromultrafine particles formed via homogeneous nucleation andcondensation processes in the vehicle exhaust (e.g. Abdul-Khalek et al., 2000) and/or combustion-generated nanopar-ticles of OC (e.g. Sgro et al., 2008). It is important to notethat we class particles formed via homogeneous nucleationshortly after emission (either in the vehicle tailpipe or inthe emission plume) as primary particles in the model, sincethey are formed from emitted precursor gases on sub-gridscales. With atmospheric dilution, semi-volatile particlesproduced via this process may undergo gas-to-particle parti-tioning; involving evaporation and possible re-condensationonto surfaces of larger particles in the exhaust plume e.g.soot or background aerosol (e.g. Zhu et al., 2002; Zhang etal., 2004). These processes make it difficult to quantify theircontribution to the average BC+OC number size distributionover the model grid box.
The appropriate emission size distribution to assume forprimary carbonaceous particles in a global model remainsambiguous. However, since the emission size distributionsused by Stier et al. (2005) and Dentener et al. (2006) arerepresentative of how the global aerosol modelling commu-nity treats the emission of carbonaceous aerosol; we use themhere in our sensitivity study. We therefore have two scenariosfor the size of BC+OC particles at emission: small particles(BCOC sm; AEROCOM, Dentener et al., 2006) and largeparticles (BCOClg; Stier et al., 2005). The emitted num-ber concentrations predicted by these two experiments canbe viewed as rough upper and lower limits to the modelledprimary BC+OC particle number concentration.
To account for sub-grid production of sulphate partic-ulates, we assume that 2.5 % of SO2 from anthropogenicand volcanic sources is emitted as sulphuric acid particles.
We use the size distribution for primary sulphate modifiedby Stier et al. (2005) from AEROCOM recommendationsfor the year 2000 (Dentener et al., 2006) (road transport:D = 60 nm, σ = 1.59; shipping, industry and power-plantemissions: 50 % atD = 150 nm,σ = 1.59 and 50 % atD =
1.5 µm, σ = 2.0; wildfire, biofuel and volcanic emissions:50 % atD = 60 nm and 50 % atD = 150 nm,σ = 1.59). Pri-mary sea spray emissions are also included and are based onGong et al. (2003).
3.3 Formation of secondary particles
A simple scheme for the formation of oxidised biogenic or-ganic compounds or secondary organic aerosol (SOA) is in-cluded in all model simulations in this study. This pro-cess involves the reaction of biogenic monoterpenes with O3,OH and NO3 (assuming the reactivity of alpha-pinene) toform a gas-phase oxidation product with a 13 % molar yield(Spracklen et al., 2006). This first stage oxidation productcan form SOA through condensing with zero vapour pres-sure onto pre-existing aerosol (Spracklen et al., 2006, 2008).Anthropogenic volatile or intermediate-volatile organic com-pounds are also known to contribute to SOA formation (e.g.Robinson et al., 2007; Henze et al., 2008), but we do not con-sider their contribution in this study.
The role of ammonium nitrate aerosol is not simulatedin GLOMAP. We recognise that the contribution of nitrateaerosol may be important for accumulation-mode particlenumber concentrations but only towards the top of the BL,where partitioning of semi-volatile gas phase species to theparticle phase occurs at reduced temperature and enhancedrelative humidity (Morgan et al., 2010). We therefore assumethat the contribution to the total particle size distribution isfairly small at the majority of the ground sites.
Secondary sulphate particles are formed through twomechanisms: binary homogeneous nucleation (BHN) ofH2SO4-H2O (Kulmala et al., 1998) to simulate nucleation inthe FT; and an empirical particle formation mechanism basedon H2SO4 specifically to capture nucleation events observedin the BL (Kulmala et al., 2006; Sihto et al., 2006). Pre-vious GLOMAP studies have shown good agreement withobservations at marine, continental and FT mountain sitesusing a combination of BHN and an empirical activation orkinetic nucleation mechanism in the BL (Spracklen et al.,2006, 2008, 2010). In Metzger et al. (2010), we tested anempirical nucleation mechanism involving low-volatility or-ganic vapour in addition to H2SO4, which showed very goodagreement for the whole vertical profile of observed particlenumber concentrations, without being restricted to the BL.
In this study, we test four nucleation mechanisms (sum-marised in Table2) intended to capture nucleation eventsobserved in the BL, while allowing BHN to occur through-out the atmosphere in all model simulations. The activationmechanism (ACT) is described by:
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12015
The sulphuric acid kinetic mechanism (KIN) is described by:
Jnuc= K[H2SO4]2 (2)
The combined organic and sulphuric acid (kinetic-type)mechanism of Metzger et al. (2010), which we call here,ORG1, is described by:
Jnuc= k [H2SO4][organic
](3)
We assume that the concentration of organic vapour ([or-ganic]) can be represented by the the gas-phase concentra-tion of the first stage oxidation product of monoterpenes (de-scribed above). We also test a new empirical mechanism ofPaasonen et al. (2010) involving kinetic-type nucleation ofsulphuric acid both homomolecularly and heteromolecularlywith low-volatility organic vapours, which we term ORG2:
Jnuc= k1[H2SO4]2+k2[H2SO4][organic
](4)
For this study, we have restricted the ACT and KIN nucle-ation mechanisms to the model BL, but allow the ORG1 andORG2 mechanisms to occur throughout the atmosphere.
The nucleation rate coefficients (see Table 2) for the ACTand KIN mechanisms have been constrained with worldwideobservations (Spracklen et al., 2010) and lie within the rangederived independently from measurements of particle forma-tion events at European ground sites (Riipinen et al., 2007).The rate coefficients for the ORG1 and ORG2 mechanismsare consistent with the studies of Metzger et al. (2010) andPaasonen et al. (2010), respectively. The value of the ratecoefficient is fixed globally in any simulation.
To take into account scavenging losses of freshly nucle-ated clusters and condensable gases during growth in the BLnucleation model simulations, the production rate of mea-sureable particles (or “apparent” nucleation rate,Japp) is con-trolled in the model by the cluster formation rate (Jnuc) andthe pre-existing particle surface area following the approxi-mation of Kerminen and Kulmala (2002):
Japp= Jnuc exp
[0.23
(1
dapp−
1
dcrit
)CS′
GR
](5)
wheredapp (nm) is the diameter of the measureable particles(here we assumedapp= 3 nm) anddcrit (nm) is the diame-ter of the critical cluster. We assumedcrit = 0.8 nm for theACT and KIN mechanisms and assume sizes of 1.5 nm and2 nm for the ORG1 and ORG2 mechanisms as used byMet-zger et al. (2010) and Paasonen et al. (2010), respectively.The growth rate of the nucleated clusters, GR (nm h−1), isassumed to be constant betweendcrit anddapp. The reducedcondensation sink, CS′ (m−2), is calculated by integratingover the aerosol size distribution (Kulmala et al., 2001). Inthe model, CS′ is calculated by summing over the aerosolsize binsj (Spracklen et al., 2006):
CS′=
∑j
βj rjNj (6)
whereβj is the transitional correction for the condensationalmass flux (Fuchs and Sutugin, 1971),rj is the particle radiusandNj is the particle number concentration. The condensa-tion sink, CS (s−1), is calculated using CS′:
CS= 4πDCCS′ (7)
whereDC is the vapour diffusion coefficient. Once a clus-ter has formed, subsequent growth in the model arises fromcondensation of sulphuric acid vapour up to a particle size of3 nm and then growth to larger sizes through the condensa-tion of both sulphuric acid and SOA (Spracklen et al., 2006).Nucleated particles are added to the model at 3 nm diameter.
3.4 Set-up of aerosol distributions
The aerosol distribution set-up used in this study has beenmodified from that used in e.g. Spracklen et al. (2006,2008, 2010) so as to track the number concentration of non-volatile (BC-containing) particle cores separately from theother species for comparison with observations. We note thatsea salt particles also contribute to the non-volatile aerosolfraction as observed at the coastal site, Mace Head (Jenningsand O’Dowd, 1990; O’Dowd and Smith, 1993). But for thisstudy, we assume the non-volatile particle number concen-tration (Dp>14 nm) measured by the DLR Falcon aircraft isdominated by primary BC (soot) particles (Rose et al., 2006;Engler et al., 2007; Birmili et al., 2009) and that the contri-bution of sea salt particles to the number concentration aloftover continental Europe is relatively small (e.g. Putaud et al.,2004).
The model was set up with two externally mixed parti-cle distributions: distribution 1 contains BC, OC and sul-phate; and distribution 2 contains sulphate, sea salt, BC, andOC. Primary BC+OC particles are emitted into distribution1 and the particles can grow by irreversible condensation ofSOA and H2SO4, with the SOA being associated with theOC component in the particles. Nucleated particles are emit-ted into distribution 2, along with primary sulphate and seaspray, but BC+OC particles enter only through coagulationwith distribution 1. The smaller particles in this distributiontend to be nucleated sulphate particles and the larger particlesare a mixture of all components.
In our previous studies, the BC+OC particles in distribu-tion 1 were moved to the equivalent size section of distri-bution 2 if they accumulated a monolayer of H2SO4 in onemodel time step – commonly referred to as a parameterisa-tion of particle ageing. Here, sulphate is allowed to accu-mulate on particles in distribution 1, and their number con-centration is depleted by coagulation with particles in bothdistributions. Both distributions are treated as hydrophilicand all particles can act as CCN and undergo wet removalprocesses.
12016 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Table 2. Summary of the GLOMAP model simulations used in this study. All model simulations include primary aerosol emissions andbinary homogeneous nucleation of H2SO4–H2O (Kulmala et al., 1998) to simulate nucleation in the FT (see Sect. 3 for details). Modelledcampaign (May 2008) mean particle number concentrations (Dp>3 nm) in the European BL (≤2000 m a.s.l.) are given for each simulation.The European domain is considered as the area between the longitudes∼65.6◦ N and∼32.1◦ N, and latitudes∼22.5◦ W and∼36.6◦ E.
# Simulation Size distribution of BL nucleation Mean particle numbername primary fossil fuel mechanism and rate concentration in the
and biofuel emissions European BL (cm−3)
1 BCOClg Large size: None 760DFF = 60 nmDBF = 150 nm(Stier et al., 2005)
2 BCOCsm Small size: None 1483DFF = 30 nmDBF = 80 nm(Dentener et al., 2006)
3 ACT-BCOClg Large size: ACT 1350DFF = 60 nm A = 2×10−6 s−1
DBF = 150 nm(Stier et al., 2005)
4 ACT-BCOCsm Small size: ACT 1871DFF = 30 nm A = 2×10−6 s−1
DBF = 80 nm(Dentener et al., 2006)
5 KIN-BCOC lg Large size: KIN 1868DFF = 60 nm K = 2×10−12cm3 s−1
DBF = 150 nm(Stier et al., 2005)
6 KIN-BCOC sm Small size: KIN 2226DFF = 30 nm K = 2×10−12cm3 s−1
DBF = 80 nm(Dentener et al., 2006)
7 ORG1-BCOClg Large size: ORG1 1967DFF = 60 nm k = 5×10−13cm3 s−1
DBF = 150 nm(Stier et al., 2005)
8 ORG1-BCOCsm Small size: ORG1 2312DFF = 30 nm k = 5×10−13cm3 s−1
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12017
3.5 Description of model simulations
The model aerosol fields were generated from an initiallyaerosol-free atmosphere initialised on 1 February 2008 andspun-up for 90 days to produce a realistic aerosol distribu-tion (Spracklen et al., 2005a). The model was set up to out-put 3-D fields every hour over a European domain. A widerange of sensitivity runs were completed to understand the ef-fect of uncertainties in the emission size of primary BC+OCparticles (Sect. 3.2) and in the mechanism and rates of BLnucleation (Sect. 3.3). The model experiments used in thisstudy are detailed in Table 2 and are split into those with andwithout BL nucleation.
4 Results and discussion
4.1 Analysis of ground site observations
In this section, we analyse surface-level aerosol measure-ments from 15 EUSAAR and GUAN ground sites (Table 1)over the EUCAARI May 2008 campaign. Summary statis-tics for total particle number concentrations (Dp>15 nm;Ntot) and number concentrations in three size ranges typicalfor CCN; Dp>50 nm (N50), >100 nm (N100), and>160 nm(N160) are given in Table 3.
4.1.1 Analysis of the monthly-mean particle sizedistribution
Figure 3 shows the May 2008 modelled number size distri-bution averaged over the 15 ground sites. The mean sizedistribution predicted by model experiments BCOCsm andBCOC lg (simulations 1–2, Table 2) is unimodal despite thebimodal emission size distribution of BC+OC particles (BFand FF emissions; Fig. 2). The primary BC+OC particles un-dergo condensation growth, coagulation, and dry/wet depo-sition after emission resulting in a modelled size distributionthat looks very different from the emitted size distribution.We are therefore not only testing the emitted size of primarycarbonaceous aerosol, but the emitted size combined withother microphysical aerosol processes in the model. The sizedistribution predicted by these experiments will also includecontributions from primary emissions of the other simulatedaerosol species (sulphate and sea salt), and secondary sul-phate particles from BHN.
Figure 4 compares the total modelled and observedcampaign-mean number size distribution at each of theground sites for all model simulations in Table 2. The gen-eral shape of the observed size distribution in the range∼80–1000 nm is well reproduced by the primary aerosol experi-ments, in particular the overlapping Aitken and accumula-tion modes typically observed at continental BL sites. Atthe majority of sites, relatively high particle concentrationswere observed in the nucleation and lower-Aitken modes.
Fig. 3. Campaign-mean modelled total number size distributionaveraged over all ground sites in Table 1. Model experiments,BCOC smand BCOClg, are described in Table 2.
Number concentrations in these size ranges are poorly cap-tured in the experiment with large primary particle emis-sions (BCOClg), resulting in a large negative bias betweenthe modelled and observed multi-site campaign-meanNtot(NMB = −69 %; m = 0.23). The overall spatial pattern ofNtot is captured well with BCOClg (R2
= 0.64). By re-ducing the emission size of the primary BC+OC particles(BCOC sm),the negative bias of the model is decreased con-siderably (NMB =−28 %; m = 0.73) and the predicted spa-tial pattern is improved further (R2 = 0.71).
Including a BL nucleation mechanism in the model (sim-ulations 3–10, Table 2) increases particle concentrations inthe nucleation and Aitken modes at the large majority ofsites, leading to better agreement with the observed size dis-tributions at small sizes. In experiment BCOClg, the meanmodelledNtot over Europe increases by a factor of∼1.6–1.9, resulting in a smaller model bias of between−53 % and−40 % depending on the BL nucleation mechanism (ACT,KIN, ORG1 or ORG2). In the BCOCsm experiment, themodel bias becomes small (range−19 to −11 %), partic-ularly with the ORG1 mechanism. When smaller primaryparticles are emitted, the increase in meanNtot over Europefrom BL nucleation is less pronounced (∼20–30 %) due tothe higher number concentration of pre-existing primary par-ticles.
The BCOCsm experiment tends to agree better with ob-servations ofNtot averaged over the IOP, suggesting highersimulated number concentrations are needed than achievedwith the BCOClg experiment, despite the large emissionsize agreeing better with measured roadside and urban BCparticle size distributions (Sect. 3.2). Including BL nucle-ation in the BCOClg experiment reduces the low bias ofthe model, but does not fully explain the shortfall inNtot.In addition, the magnitude of the slope of the linear regres-sion between modelled and observedNtot remains low (m=
0.22–0.26) and there is a decrease in the spatial correlationbetween model and observations with the ORG1 (R2
12018 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Fig. 4. Campaign-mean simulated (colour) and observed (black) total number size distributions at each ground site. Model experimentslisted in the legend are described in Table 2.
and ORG2 (R2 = 0.59) mechanisms. These results suggestpossible errors in the modelling of nucleation events (dis-cussed in Sect. 4.4), which may be a reason why BL nucle-ation is unable to explain the shortfall.
The dependence of modelled concentrations on the as-sumed size of the primary particles decreases with the size ofparticles being considered. For example, the mean modelledN50 increases by∼60 % in the European BL between theBCOC lg and BCOCsm experiments, whileN100 andN160increase by∼45 % and∼20 %, respectively. The model sim-ulations without BL nucleation compare well with the obser-vations ofN50, N100 andN160 (Table 3), confirming that theunderpediction ofNtot is largely due to an underprediction ofnumber concentrations in the range 15–50 nm (N<50). Fig-ure 5 shows the normalised mean bias between hourly-meanmodelled and observedN<50 andN50 (NMBhourly) at eachsite for the IOP. The spatial pattern ofN50, N100, andN160over Europe is captured well by the model (R2
= 0.47–0.86).
When we assume a small initial size for primary BC+OCparticles (BCOCsm), we find good agreement with sur-face observations ofN50 (NMB = −18 %, m = 0.80) andN100 (NMB = −1 %,m = 0.81) averaged over the IOP. Withthe BCOClg experiment the model is biased low forN50(NMB = −52 %, m = 0.44) andN100 (NMB = −29 %, m =
0.64). For N160, the model bias is small in experimentBCOC sm (NMB = 9 %,m = 0.65), but in contrast to com-parisons with observedN100, N50 and Ntot, we find thebest agreement with observedN160 over the IOP is with theBCOC lg experiment (NMB =−1%,m = 0.74).
Including BL nucleation in the model increases the cam-paign meanN50 andN100 in the European BL by 23–36 %and 14–20 % respectively in the BCOClg experiment, andby 8–12 % and 5–8 % respectively in the BCOCsm exper-iment. The increases in particle number concentrations de-pend on the nucleation mechanism (the smallest increase inN50 and N100 is achieved with the ACT mechanism; the
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12019
Table 3. Summary statistics for total particle number concentrations (Dp>15 nm;Ntot) and for concentrations of particles in three size-ranges typical for CCN;Dp>50 nm (N50), >100 nm (N100) and>160 nm (N160). The normalised mean bias (NMB), slope of the linearregression (m) and correlation coefficient (R2) are calculated between the simulated and observed campaign-mean number concentrations ateach ground site.
largest with the ORG1 mechanism). These results are similarto the mean enhancements to CCN found by Spracklenet al.(2008); CCN number concentrations at 1% and 0.2% super-saturation (CCN (0.2 %)) were found to increase by 30 % and6–15 % respectively at European ground sites. Pierce andAdams (2009) also show a∼5% increase in BL CCN (0.2 %)over Europe when activation BL nucleation is included intheir model.
The impact of BL nucleation on CCN-sized particle num-ber concentrations is considerably smaller than forNtot(given above) and for the total particle number concentra-tion with Dp>3 nm (see Table 2, column 5). The dampenedresponse ofN50 andN100 to BL nucleation arises from anincrease in coagulation and condensation sinks from an ad-ditional source of secondary particles, thereby reducing thesurvival probability of ultrafine particles and reducing thecondensational growth of these particles to CCN sizes (e.g.Pierce and Adams, 2007; Kuang et al., 2009).
Including BL nucleation in the BCOClg experiment re-duces the negative model bias inN50 andN100; the smallestbias in bothN50 (−37 %) andN100 (−17 %) is achieved withthe ORG1 mechanism. In the BCOCsmexperiment, the biasin N50 is also reduced by including BL nucleation; the small-est bias (−11 %) is achieved with the KIN and ORG1 mech-anisms. However, forN100 all nucleation mechanisms leadto a slightly larger model bias (although the NMB remainssmaller than 10 %). The impact of BL nucleation onN160is fairly negligible (increasing mean concentrations over Eu-rope by less than 1 %), resulting in small changes in themodel bias in this size range.
When BL nucleation is included, there is little improve-ment (if any) in the slope of the linear regression and corre-lation coefficient between simulated and observed multi-sitecampaign-meanN50, N100, andN160. Without further sup-
porting evidence, these results would suggest that the modelis able to explain the observed number concentrations ofCCN-sized particles averaged over the IOP reasonably well,without the need for BL nucleation, if a small initial size isassumed for emitted BC+OC particles.
4.1.2 T-statistics at each ground site
The NMB between modelled and observed multi-sitecampaign-mean number concentrations can be misleading ifthere is cancellation of positive and negative biases at dif-ferent ground sites or if day to day variability is poorly simu-lated. To overcome the possibility of a cancellation of biases,we have analysed the statistical significance of the differ-ence between the model and the observations at each groundsite using the hourly data. Here, we include an analysis ofN<50, since the underprediction ofNtot with the BCOClgandBCOC smexperiments is largely due to an underpredic-tion of number concentrations at the small end of the sizedistribution.
For this analysis, we calculated a pairedt-test of the hourlytime series of particle concentrations in the different sizewindows and calculated the significance at the 99 % confi-dence level. To take into account temporal correlation in themodelled and observed time series we adjusted thet-statisticby calculating an “effective sample size” for each site, usingthe method of Wilks (1997) for second order autoregressive(AR(2)) data. We found the hourly time series were best fitwith an AR(2) process, using a Durbin-Watson test (Durbinand Watson, 1950) to examine the residuals of the series. TheAR(2) process best accounted for the diurnal variability andrandom variations visible in the observed and modelled time-series.
12020 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Fig. 5. Normalised mean bias (NMB) between hourly-mean modelled and observed particle number concentrations at each ground site.NMB is shown for model experiments 1–4, Table 2 for number concentrations in the size ranges;(a) Dp = 15–50 nm and(b) Dp>50 nm.
In this section, we essentially test the significance of allthe plausible primary aerosol experiments and BL nucleationexperiments and so group the model simulations into thosewithout BL nucleation (simulations 1–2, Table 2) and thosewith BL nucleation (simulations 3–10, Table 2). The rangein the first set of experiments represents the uncertainty inthe assumed emission size distribution for primary BC+OCand the range in the second set of experiments represents theuncertainty in the empirical BL nucleation parameterisationused in the model. The results of the significance tests aresummarised in Fig. 6.
ForN<50, we find that without BL nucleation, the model-observation difference is statistically significant at all of theground sites. Figure 5a shows that at 12 of the 15 sites theNMBhourly is fairly large and negative (BCOClg, range−98to −83 %; BCOCsm, range−77 to −33 %). The excep-tions are at Cabauw and Finokalia where the modelledN<50spans the observations (concentrations are underpredictedwith BCOC lg and overpredicted with BCOCsm), and atthe high altitude site, Jungfraujoch, where the meanN<50is overpredicted by a factor of∼2.0. This overprediction atJungfraujoch was also found in our global analysis of particlenumber concentrations (Spracklen et al., 2010). When someform of BL nucleation is included, the model-observationdifference becomes insignificant at 6 sites, showing that, sta-tistically, nucleation is an important process affectingN<50at at least 40 % of the ground sites.
For N50, the model-observation difference is statisticallysignificant at 12 of the 15 sites without BL nucleation. At the3 sites where the model-observation difference is insignifi-cant (Jungfraujoch, Melpitz and Cabauw), it is the BCOCsmexperiment that captures the observations. The observationsat Jungfraujoch are also captured with the BCOClg exper-
iment. At these 3 sites, the NMBhourly is very small (range−7 to 5 %). But at the remaining 12 sites (with a significantdifference) the model bias is still fairly small (Fig. 5b): for all12 sites the bias is smallest with the BCOCsm experiment(between−43 % and 21 %).
When some form of BL nucleation is included, the model-observation difference inN50 becomes insignificant at anadditional 7 sites. For these sites, BL nucleation makesan important contribution toN50. At Jungfraujoch (wherethe difference was insignificant with experiments BCOCsmand BCOClg), including BL nucleation increases the modelbias, but at the 99 % confidence level the model-observationdifference remains statistically insignificant. Overall, withBL nucleation the difference between modelled and observedN50 is insignificant at two thirds of the ground sites. Thus,the model with BL nucleation is in better agreement with theobservations than the model without BL nucleation.
For N100, we find that at 12 sites there is a statisti-cally significant difference between the model and obser-vations in experiments without BL nucleation. At the 3sites where model-observation difference is statistically in-significant, again it is the BCOCsm experiment that cap-tures the observations. This is the same proportion of sitesas for N50, but at the sites where the difference is signif-icant the NMBhourly is generally smaller forN100. For 9sites the bias is smallest with the BCOCsm experiment (be-tween−19 % and 18 %), and for 2 sites the bias is smallestwith the BCOClg experiment (−32 % at Cabauw and 9 %at Finokalia). At 1 site (Jungfraujoch), there is a large neg-ative bias with both model experiments (BCOCsm,−69 %;BCOC lg, −81 %).
When BL nucleation is included, the model-observationdifference inN100 is no longer significant at an additional
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12021
Fig. 6. Statistical significance between hourly-mean modelledand observed particle number concentrations in three size ranges;Dp = 15–50 nm (N<50), Dp>50 nm (N50), and Dp>100 nm(N100). The red dots show site locations where the difference be-tween the model and observations is statistically significant at the99 % confidence level; the black dots show the locations where thedifference is insignificant.(a), (b) and(c) show results for modelexperiments without BL nucleation (1–2, Table 2);(d), (e) and(f)show results for the experiments including BL nucleation (3–10,Table 2).
4 sites (Hyytiala, Vavihill, Monte Cimone and Aspvreten).However, at 1 of the 3 sites where the difference was insignif-icant with experiment BCOCsm (Schauinsland), addingBL nucleation results in an overprediction ofN100 and themodel-observation difference becomes significant. In total,the model with BL nucleation is able to capture the observa-tions at almost half of the ground sites.
We conclude from these time series comparisons that fornumber concentrations at the small end of the size distribu-tion, N<50, we need to include BL particle formation for thedifference between model and observations to be statisticallyinsignificant at roughly half of the ground sites. It is possi-ble that a larger contribution from BL nucleation is neededin the model to capture the observations at some of the re-maining sites. The observedN<50 may also be influenced bylocal sources, particularly at the more polluted sites (Ispra,Cabauw and Melpitz), or by diurnal cycles in aerosol at themountain sites Jungfraujoch (Weingartner et al., 1999) andPuy de Dome (Venzac et al., 2009), that the model is unableto capture due to its fairly coarse resolution.
The results of thet-tests show that the model with BLnucleation also gives the best overall agreement with obser-vations ofN50 andN100, capturing the observations at twothirds and almost half the sites respectively. However, if wetake into account the±10 % uncertainty of the S/DMPS mea-surements (Wiedensohler et al., 2010),N50 andN100 can beexplained at almost all of these sites without the need for
BL nucleation. In total, the difference between the modelwithout BL nucleation and observations (±10 %) is statisti-cally insignificant at 8 sites forN50 and 10 sites forN100.Including BL nucleation in the model, the observations canbe captured within±10 % at an additional 4 sites forN50(Hyytiala, Mace Head, Vavihill, and Schauinsland) and anadditional 2 sites forN100 (Hohenpeissenberg and Kosetice).Therefore at the majority of ground sites, it is difficult to de-tect the contribution of BL nucleation toN50 andN100 withinthe uncertainty of the observations.
If we adjust the interval of the modelled and observed timeseries to better represent the average residence time of airin the model grid box (∼5–20 h), the results of the signif-icance tests are improved but the conclusions regarding BLnucleation remain unchanged. If we compare the model toa 20-h running average of the measurements, the number ofsites where the difference between modelled and observedN<50 is statistically insignificant is increased to 12 out of15 sites, but at all but 2 of these sites it is still necessaryto include BL nucleation to capture the observations. ForN50 andN100 the number of sites with an insignificant dif-ference is increased to 12 and 13 sites respectively, but BLnucleation is only needed to capture the observations at≤2of these sites. These results confirm the conclusions fromthe hourly time series analysis; to capture ground-based ob-servations ofN<50 we need to include BL nucleation in themodel, but for CCN-size number concentrations only a fairlysmall contribution from BL nucleation (if any) is needed tocapture the observations.
We recognise that BL nucleation may be important forN<50, N50 andN100 at more than the number of sites dis-cussed above, but that the observed nucleation events maynot be adequately modelled for this period by the mecha-nisms applied in this study (Sect. 4.4). The sites at whichBL nucleation is needed in the model to capture the hourlyobservations ofN<50, N50, andN100 are summarised in Ta-ble 4.
4.1.3 Analysis of particle concentration frequencydistributions
Normalised histograms of the frequency distribution of mod-elled and observedN50 are shown for each site in Fig. 7. Asin Gilardoni et al. (2011), we calculate the degree of overlapbetween the modelled and observed frequency distributions(given in Fig. 7). Figure 7 shows there is some dependenceof the best-fit assumption of BC+OC particle emission sizeon site location. For example, at Jungfraujoch the range ofobserved concentrations is captured best when larger primaryBC+OC particles are emitted, with a distribution overlap of78 %. But at all other sites, the BCOClg experiment not onlyunderpredictsN50, but also underpredicts the range of con-centrations observed (average overlap of 42 %). The rangeof observedN50 is captured much better at most sites when
12022 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Table 4. Summary of the ground sites at which a statisti-cal improvement in the predicted hourly-mean number concen-tration is achieved by including BL nucleation in the model.The results are given for particle number concentrations inthree size ranges;Dp = 15–50 nm (N<50), Dp>50 nm (N50), andDp>100 nm (N100). The “+” sign indicates where a statisticallysignificant difference between the model and observations is re-moved by including BL nucleation. The “–” sign indicates wherethe reverse occurs i.e. including BL nucleation leads to a statisticallysignificant overprediction of the observedN<50, N50, or N100. “0”indicates where there is no statistically significant change in the pre-dicted particle number concentrations with BL nucleation.
smallerBC+OC particles are emitted in the model (averageoverlap of 67 %).
Including BL nucleation increases the range of simulatedN50 in experiment BCOClg and improves the agreement be-tween modelled and observed frequency distributions (aver-age overlap of 53–56 %, versus 42 % without BL nucleation).The impact of BL nucleation is fairly small on the range ofN50 predicted by experiment BCOCsm(average overlap of67–68 % versus 67 % without BL nucleation), and at 7 sitesthe distribution overlap is decreased slightly. At Finokaliaand Hyytiala the distribution overlap becomes greater in ex-periment BCOClg than in BCOCsm. However, at twothirds of the sites, the range of observedN50 is captured bestwith experiment BCOCsm(with or without BL nucleation).
In general, we find that the assumption of smaller BC+OCparticles with higher number concentration gives the bestagreement with the observed frequency distribution ofN50.However, if we include BL nucleation in the model, a num-ber of sites fit better when we assume larger emitted particles.The dependence of the best-fit model on location suggests ei-ther that the emitted primary particle size/number concentra-tion is more variable across Europe than assumed by the con-stant emission size distribution prescribed in the model, orthat atmospheric processes (including BL nucleation) might
be influencing the shape of the size distribution in ways notrepresented in the model.
4.2 Supporting aircraft observations in the boundarylayer
Figure 8 shows the mean vertical profiles of particle num-ber concentrations measured by the DLR Falcon aircraft overthe IOP in three size ranges; 4–10 nm (N4−10), 10–160 nm(N10−160), and 160–1040 nm (N160−1040). These size rangesare representative of the nucleation, Aitken, and accumula-tion mode size classes, respectively. Comparison with theprimary aerosol model experiments (1–2, Table 2) in the BL(<2 km a.s.l.) is consistent with the analysis of the groundsite data. There is good agreement at the larger sizes but anincreasing model underprediction of particle concentrationsat the small end of the size distribution. Summary statisticsfor N4−10, N10−160, andN160−1040are given in Table 5.
It is important to note thatN4−10 is a challenging quantityto compare the model with because nucleation mode particlesoften appear as distinct events in the aircraft data time series.Where there are no nucleation mode particles observed, mea-surements ofN4−10 can be negative i.e. from when measure-ments ofN10 are larger thanN4, indicating some uncertaintyin the observations. For 3 flights, the meanN4−10 in the BLis negative, resulting in low correlation between mean mod-elled and observed number concentrations in this size range(R2
≤0.01 with all model simulations). For this reason, wefocus mainly on the measurements at larger size (N10−160andN160−1040) and only show the NMB forN4−10 in Ta-ble 5.
Between altitudes of∼2.5 and 5 km the mean simu-lated concentrations of nucleation, Aitken, and accumulationmode particles agree reasonably well with the aircraft obser-vations, and generally remain within∼1σ of the observations(Fig. 8). However, the model is unable to capture the peak inmeanN4−10 andN10−160observed in the BL.
Without BL nucleation, the model predicts very few nu-cleation mode particles in the BL, resulting in substantialunderprediction ofN4−10 (NMB = −100 %). Including BLnucleation in experiment BCOClg results in a considerabledecrease in the model bias with the KIN (NMB =−38 %) andORG2 (NMB =−33 %) mechanisms. This is due to a largeincrease in the mean simulatedN4−10 for 3 flights, caus-ing the model to overpredict observed BL concentrationsby up to a factor of 9 for these flights, although the meanN4−10 is still underpredicted substantially for the majority offlights with these two simulations. In the BCOCsm exper-iment, the bias is only reduced slightly with BL nucleation(NMB = −96–−80 %, depending on the mechanism). In thevertical profile ofN4−10 (Fig. 8a), the experiments with BLnucleation due to biogenic precursors predict the highest con-centrations between∼4 and 5 km since the ORG1 and ORG2mechanisms are not restricted to the BL (unlike the KIN andACT mechanisms) and can occur throughout the atmosphere
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12023
Fig. 7. Normalised histograms of the frequency distribution of hourly-mean simulated (colour) and observed (black) number concentrationsof particles withDp>50 nm (N50) for May 2008 at each ground site. Bin size depends on the maximumN50 observed at each site; numberconcentrations are divided into 15 equally spaced bins. The percentage overlap between the modelled and observed frequency distributionsis given in the top right hand corner. Please note that the colours representing the model experiments in this figure (chosen to be easilydistinguishable from each other) differ from the previous and following figures. Model experiments are described in Table 2.
providing the concentration of organic vapour is sufficientlyhigh.
With experiment BCOClg the model bias is also largefor N10−160 (NMB = −85 %, m = 0.04), underpredictingmean concentrations for every flight by a factor of between2.4 and 11.9. When smaller primary particles are emitted(BCOC sm) the bias inN10−160 is reduced (NMB =−63 %,m = 0.17), but mean concentrations are still underpredictedfor every flight by a factor of between 1.4 and 4.9. Thespatial distribution ofN10−160 is also improved by emittingsmaller BC+OC particles (BCOClg, R2
= 0.04; BCOCsm,R2
= 0.21). Including BL nucleation reduces the bias fur-ther to −59–−44 % (depending on the mechanism and on
the BC+OC emission size). However, the smaller NMB ismainly due to a large increase in modelled concentrationsand overprediction for 1 flight (NMB with this flight removedis also shown in Table 5). As a result, the spatial distribu-tion of N10−160 is not as well captured with BL nucleation(R2<0.03).
The whole vertical profile ofN160−1040is captured fairlywell by the model, with a peak in concentration in theBL that rapidly decreases above an altitude of∼2.5 km(Fig. 8c). The model without BL nucleation slightly underes-timates the meanN160−1040observed in the BL (BCOCsm,NMB = −22 %, m = 0.31; BCOClg, NMB = −21 %, m =
0.34). Although this particle size-range is generally
12024 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Fig. 8. Vertical profiles of observed (black) and modelled (colour) particle number concentrations in the diameter ranges:(a) 4–10 nm,(b) 10–160 nm, and(c) 160–1040 nm. Observations are from the DLR Falcon 20 aircraft. The average over all measurement flights performedduring the LONGREX campaign (May 2008) is shown (sectioned into 600 m altitude bins). The error bars and shading represent the standarddeviation of the model and observations, respectively. Model experiments listed in the legend are described in Table 2.
dominated by secondary aerosol mass, the number concen-trations may well be explained by primary emissions (andsome contribution from BHN) because the condensation ofsecondary aerosol species onto primary particle cores occurswithout a change in number concentration. Particle growthvia condensation of H2SO4 and SOA is included in all modelexperiments. There is also likely to be some contributionto growth from condensation of ammonium nitrate (not in-cluded in the model) towards the top of the BL (Morgan et al.,2010, see Sect. 3.3), which may explain some of the modelunderprediction ofN160−1040.
The impact of BL nucleation on number concentrations atthe large end of the size distribution is relatively small, withan average change in the mean simulatedN160−1040for eachflight of ∼2 %. But in the BCOClg experiment, the overallagreement between mean modelled and observedN160−1040is generally improved with BL nucleation. This can be inter-preted as a decreasing influence of primary emissions aloftin the BL compared with observations at the surface. Fornumber concentrations in this size range the BCOClg ex-periment gives slightly better agreement with the aircraftobservations which is consistent with comparisons with theground-based observations.
4.3 Supporting observations of non-volatile particles
Further information on the number concentrations of car-bonaceous particles can be obtained from measurements ofnon-volatile cores. Here, we use the measurements of non-volatile particle size and number concentration made on-board the DLR Falcon aircraft. Previous studies using mea-surements of non-volatile particles have found that the sub-micron non-volatile fraction essentially consists of primaryBC (soot) particles from combustion sources with some con-tribution from organic compounds (Rose et al., 2006; En-gler et al., 2007; Birmili et al., 2009). Most of the volatileaerosol species such as sulphate, nitrate, ammonium andsome volatile organic compounds are evaporated at tempera-tures below 250◦C. Other non-volatile aerosol species suchas inorganic salts and crustal material are likely to contributemainly to measured non-volatile number concentrations inthe super-micron size range (Rose et al., 2006; Birmili et al.,2009). We therefore assume that the observed submicronnon-volatile particles can be compared with the simulatedBC+OC particle number concentration from the model.
Figure 9 shows a mean campaign vertical profileof observed non-volatile particle number concentration
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12025
Table 5. Summary statistics for particle number concentrations in the diameter ranges; 4–10 nm (N4−10), 10–160 nm (N10−160), and 160–1040 nm (N160−1040). The normalised mean bias (NMB), slope of the linear regression (m) and correlation coefficient (R2) are calculatedbetween the simulated and observed mean number concentrations in the BL (≤2000 m a.s.l.) for each flight performed by the DLR Falconaircraft during LONGREX, May 2008.R2 andm are not given forN4−10 for reasons explained in Sect. 4.2. Values in brackets are statisticscalculated with 1 flight (the second flight on 22 May) removed.
(Dp>14 nm) measured using a thermodenuder and CPC,compared with the modelled number concentration ofBC+OC particles (Dp>14 nm). The highest non-volatile par-ticle concentrations were observed in the BL as a result ofsurface primary emissions. The model captures the generalshape of the observed vertical profile with maximum numberconcentrations in the BL decreasing with increasing altitude.
Figure 9 shows how the size of emitted BC+OC parti-cles affects the number concentration for fixed mass (sim-ulations 1–2, Table 2). On average, there is a factor∼3.8change in total simulated BC+OC particle number concen-tration (Dp>3 nm) in the European BL between experimentsBCOC lg and BCOCsm. We note that this ratio is not thesame as the ratio of emitted number concentrations (which isa factor of∼ 4.4 for FF emissions) due to non-linear effectsof microphysical and removal processes on particle concen-trations.
The model underpredicts the mean non-volatile particleconcentrations in the BL for every flight by a factor of 2.5–10.4 (NMB =−78%), when we assume emission of largeBC+OC particles (BCOClg). Emitting smaller BC+OCparticles (BCOCsm) reduces the bias (NMB =−32 %)and mean concentrations are predicted within a factor of2.8. To achieve good agreement with the observations(NMB = −3 %) we need to further reduce the assumed emis-sion size of carbonaceous aerosol in the model by a factor of∼1.2 (experiment “BCOCvsm” in Fig. 9), which increasesthe total simulated BC+OC number concentration over Eu-rope by a factor of∼1.5 relative to BCOCsm.
The model underprediction of the BL non-volatile particlenumber concentration is largest in the BCOClg experiment,in which the emitted BC+OC size distribution is more in-linewith measured BC emission sizes in the literature (Sect. 3.2).Only by increasing the emitted number concentration of car-bonaceous aerosol in BCOClg by more than a factor of∼6
(in experiment BCOCvsm) are we able to capture the ob-servations, which suggests that (i) the removal of BC+OCparticles is too efficient in the model, (ii) the non-volatilecounter is not measuring the same particles as assumed inthe model, or (iii) the model is missing a large contribu-tion of non-volatile particles. Concerning (i): if we substan-tially reduce the in-cloud nucleation scavenging efficiencyin the model (by decreasing the fraction of condensate thatis converted to dynamic rain in 6 h by a factor of 10), thecampaign-mean BC+OC number concentration (Dp>14 nm)is increased by only∼20 % in the BL. Concerning (ii): it ispossible that pyrolysis of volatile OC in the thermodenudermight produce a residual core (a few nanometers in diame-ter) which appears as a non-volatile particle. If the residualsproduced by this process are larger than 14 nm they wouldbe counted by the CPC, resulting in an overestimation of theambient non-volatile particle number concentration.
Concerning (iii) (and (ii)): in experiments BCOCsm andlg it is possible that we are neglecting some contribution
from residuals of partly volatile species, detected as a non-volatile size mode withDp<20 nm in an urban environment(Birmili et al., 2010) and in the rural background (Engleret al., 2007); the composition of which is unknown. Birmiliet al. (2010) suggest the non-volatile residuals originate fromparticles containing a high volume fraction of volatile species(∼90 %), such as organic compounds from both direct vehi-cle emission and secondary formation processes. Secondaryparticles withDp<20 nm have also been observed to con-tain non-volatile residuals at rural sites (Wehner et al., 2005;Ehn et al., 2007). In the model we treat all secondary (nucle-ated) particles as volatile. It is unclear whether these residu-als make an important contribution to the non-volatile parti-cle number concentrations observed aloft in the BL, but theycould explain some of the model discrepancy.
12026 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Fig. 9. Vertical profile of measured non-volatile particle numberconcentration (Dp>14 nm) from the DLR Falcon 20 aircraft (black)compared with modelled number concentration of BC+OC particles(Dp>14 nm) (colour). The average over all measurement flightsperformed during the LONGREX campaign (May 2008) is shown(sectioned into 600 m altitude bins). The error bars and shading rep-resent the standard deviation of the model and observations respec-tively. Model experiments, BCOCsm and BCOClg, are describedin Table 2. In the BCOCvsm experiment, the peak diameter ofthe BC+OC emission size distribution in BCOCsm is reduced bya factor of∼1.2.
Assuming a very small initial size for primary BC+OCparticles (experiment BCOCvsm) increases the modellednon-volatile particle number concentration, compensatingfor non-volatile residuals that may be neglected by the simplerepresentation of non-volatile particles in our model. How-ever, we believe a diameter of 25 nm assumed forDFF inexperiment BCOCvsm, is unrealistically small for directlyemitted BC particles from traffic sources and (taking into ac-count sub-grid scale processing) this diameter may also betoo small for the mean BC+OC particle size over a largemodel grid box (Sect. 3.2).
We also compare the model with the non-volatile numbersize distribution in the dry diameter range∼0.265–2.25 µmmeasured by the OPC instrument. Seven hours of measure-ments were selected (from 6 different flights), where thehourly-mean altitude of the aircraft was lower than 2 km a.s.l.Figure 10 shows the simulated number size distribution ofBC+OC particles compared with the observed non-volatileparticle number size distribution in the BL. We assume theevaporation of all volatile species occurs before measure-ment, so that the observed size distribution in Fig. 10 showsthe size distribution of non-volatile particle cores. We try toreplicate this in the model by calculating the size distributionof the BC particle “cores”. Sulphate and SOA that have accu-mulated on the BC+OC particles in distribution 1 during theageing process act to increase the particle size (Sect. 3.4). Weremove this effect by calculating the size of the BC particlesfrom the mass of BC and the BC+OC particle number con-
Fig. 10. Mean number size distributions of measured non-volatileparticles (black) and of modelled carbonaceous particles (colour)for all flight hours with a mean altitude less than 2 km a.s.l. The to-tal modelled size distribution of BC particle cores is shown in boldand the modelled size distribution of aged BC+OC (with condensedSO4 and SOA) is shown for sizes larger than∼100 nm. The er-ror bars and shading represent the standard deviation of the modeland observations, respectively. Model experiments, BCOClg andBCOC sm, are described in Table 2.
centration (assuming a density of 1.8 gcm−3). Figure 10 alsoshows the modelled size distribution of all components indistribution 1; a mixture of fresh and aged carbonaceous par-ticles (SO4/BC/OC). The modelled size distribution of BCparticle cores is shifted to smaller sizes compared with themodelled distribution of aged BC+OC particles.
The observed mean non-volatile particle number size dis-tribution is underpredicted by the modelled “BC-only” sizedistribution, but the agreement between the modelled distri-bution of aged BC+OC particles and the observations is rea-sonably good (Fig. 10). If we compare the modelled andobserved non-volatile particle number concentrations in thesubmicron size range (between∼0.265 and∼1 µm), the dif-ference between the integrated flight-mean modelled agedBC+OC particle distribution and the observations is statis-tically insignificant at the 95 % confidence level. The dif-ference between the submicron modelled BC-only size dis-tribution and the observations, on the other hand, is statis-tically significant. These results suggest that the measurednon-volatile particle size distribution does not only consistof BC, but is likely to include contributions from non-volatileorganic matter. In addition, sea salt particles (Jennings andO’Dowd, 1990; O’Dowd and Smith, 1993) and mineral dustmay contribute to the measured non-volatile particle numberconcentrations, which are not included in the modelled sizedistribution in Fig. 10. However, these species are only likelyto make substantial contributions in the super-micron sizerange. At these large sizes the differences between modelexperiments BCOClg and BCOCsm are relatively small.
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12027
Fig. 11. May 2008 time-series of hourly-mean modelled (colour) and observed (black) particle number concentrations (Dp>15 nm) at eachground site. Model experiments listed in the legend are described in Table 2.
4.4 Time series of particle number concentrations
Conclusions regarding the best nucleation mechanism arehard to draw because of the difficulty in detecting a statis-tically significant impact of BL nucleation on CCN-sizedparticle number concentrations within the uncertainty in theground based observations (Sect. 4.1.2). In addition, the pre-dicted time series ofNtot has a temporal pattern that is inpoor agreement with the observations (Fig. 11). The corre-lation between modelled and observed hourly-meanNtot andN100at each site are given in Table 6. Without BL nucleation,the correlation between hourly-mean modelled and observedNtot is fairly low at most of the sites (averageR2
hourly= 0.08),but is reduced further when BL nucleation is included (av-erageR2
hourly = 0.05). The exception is at Cabauw wherethe correlation between model and observations is fairlygood with all simulations (averageR2
hourly= 0.28). The poor
model skill is reflected in theR2 values in Table 3: the corre-lation between modelled and observed campaign-meanNtotis generally reduced when BL nucleation is included.
Increasing the 1-h interval of the time series to 20 h (an es-timate of the average residence time of air in the model gridbox), marginally improves the correlation between modelledand observedNtot (averageR2
hourly= 0.17 without BL nucle-
ation; averageR2hourly = 0.10 with BL nucleation). The tem-
poral correlation between the model with BL nucleation andobservations remains lower than the model without BL nu-cleation, suggesting possible errors present in the modellingof BL nucleation events for this period.
Periodic features visible in the simulated time series ofNtot at some of the sites (Fig. 11) result from the develop-ment of the model BL. These features are most prominentduring Period A at the relatively low level, Central Euro-pean sites Melpitz, K-pustza, Ispra, and Cabauw, where theinfluence of BL nucleation on modelledNtot is relativelysmall. At night-time the model BL becomes shallower andstably-stratified increasing particle number concentrations atthe surface, which decrease through the day-time (in the ab-sence of BL nucleation) as the model BL height increases.
12028 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Table 6. Correlation coefficient (R2) between observed and simulated hourly mean particle number concentrations(a) Dp>15 nm and(b) Dp>100 nm at each ground site. Ground site acronyms are given in Table 1.
(a) Model ASP CBW FKL HPB HTL JFJ JRC KPO KTC MHD MPZ MTC PDD SLD VHLexperiment
Analysis of modelled and observed condensation sink(CS; Eq. 7) at all sites suggests that the nucleation sinkterm is not the cause of the poor agreement (Fig. 12).The statistical values for CS with all model experiments(NMB = −29–6 %,m = 0.53–0.69,R2
= 0.73–0.74, averageR2
hourly= 0.25) are considerably better than forNtot. A morelikely reason for the poor prediction of nucleation events issulphuric acid. Figure13 compares the simulated time se-ries of gas-phase sulphuric acid concentrations with chemicalionization mass spectrometer (e.g. Berresheim et al., 2000)measurements at Melpitz.
All model simulations underpredict the high concentra-tions of sulphuric acid observed at Melpitz during PeriodA by a factor of 1.7–4.6. Lower concentrations observedbetween∼18–24 May are likely to have contributed to thedecrease inNtot during Period B (Fig. 11), despite the re-duction in observed CS. In contrast, the modelled concentra-tions of sulphuric acid increase in Period B at Melpitz andat the majority of ground sites, driving the increase in nu-cleation events predicted by the model. One explanation forthe relatively poor agreement is that processes that drive day-to-day changes and hourly variability in gas-phase sulphuricacid concentrations are unaccounted for in the model. In par-ticular, we neglect the impact of cloud cover on incomingradiation and OH concentrations. If the dynamic of the di-urnal cycles of sulphuric acid concentrations is wrong in themodel then this can result in too small nucleation rates at thesurface.
The simplified SOA formation scheme used in the modelmay also be responsible for the relatively poor correlationbetween modelled and measured hourly means. In particu-lar, the scheme does not include contributions from anthro-pogenic volatile or intermediate-volatile organic compounds,which may have large implications for the growth rate andsurvival of the particles formed by BL nucleation. We recog-nise that more attention to modelling gas-phase H2SO4 andSOA formation is needed in future studies.
5 Summary and conclusions
We have evaluated the global aerosol microphysics model,GLOMAP against extensive measurements of total parti-cle number concentration and size distribution made duringthe EUCAARI May 2008 campaign. We have focused onaerosol concentrations in the European boundary layer (BL),using surface-based measurements from 15 EUSAAR andGUAN ground sites with airborne measurements from theDLR Falcon 20 aircraft.
The aim of this study was to better understand how pri-mary particle emissions and secondary particle formation inthe BL influence total particle number concentrations overEurope, and how the influence varies across the particlesize distribution (nucleation, Aitken and accumulation modesizes). We have extended the monthly-mean analysis of to-tal particle concentrations in Spracklen et al. (2010) to in-clude aerosol measurements from different platforms, higher
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12029
Fig. 12. May 2008 time-series of hourly-mean simulated (colour) and observed (black) condensation sink (CS) at each ground site. Modelexperiments, BCOClg and BCOCsm, are described in Table 2.
Fig. 13. Time-series of simulated (colour) and measured (black)concentrations of gas-phase sulphuric acid at the Melpitz groundsite for May 2008. Model experiments, BCOClg and BCOCsm,aredescribed in Table 2.
temporal resolution, additional nucleation mechanisms, andadditional aerosol measurements such as the size distributionand non-volatile particles. This analysis was a demandingtest for a global model, comparing with spatially and tempo-rally intensive observations in a special meteorological situ-ation over a relatively short period. During the campaign pe-riod, Central Europe was almost entirely influenced by east-erly flow, which is not the most usual case.
We found that for the campaign period, the model was ableto capture the mean particle number size distribution overEurope well for particle sizes relevant for CCN (Dp>50 nm)without the need for BL nucleation. The spatial distributionsof campaign-mean number concentrations larger than 50 nm(N50) and 100 nm (N100) dry diameter were well capturedat the ground sites (R2&0.8). In addition, the normalisedmean bias (NMB) between mean modelled and observedN50(−18 %) andN100 (−1 %) was small if we assumed a smallsize for emissions of BC+OC particles, as used by AERO-COM (Dentener et al., 2006).
The mean number size distribution at sizes smaller than50 nm diameter (N<50) was generally underpredicted inmodel experiments without BL nucleation. The difference
12030 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
between modelled and observedN<50 was found to be sta-tistically significant at all ground sites. The average overlapof modelled and observed frequency distributions ofN50 andN100 was over 65 % (with small primary BC+OC particles),but less than 55 % forN<50 without BL nucleation.
Comparisons with particle number concentrations roughlyin the nucleation, Aitken, and accumulation mode size rangesmeasured by the DLR Falcon aircraft in the BL were consis-tent with the analysis of the ground station data. We foundgood agreement at the larger sizes but an increasing modelunderprediction of particle concentrations at the small end ofthe size distribution.
We tested four empirical parameterisations for secondaryparticle formation in the BL: the activation (ACT) and kinetic(KIN) mechanisms where the cluster formation rate is pro-portional to the gas-phase sulphuric acid concentration to thepower 1 or 2, respectively, and two newly developed mecha-nisms (ORG1 and ORG2) where the formation rate dependson the concentration low-volatility organic vapours in addi-tion to sulphuric acid.
When BL nucleation was included in the model, the shapeof the predicted campaign-mean size distribution at sizes<50 nm was improved considerably, and the negative biasin N<50 was reduced. The difference between modelled andobservedN<50 became statistically insignificant at roughlyhalf of the ground sites with BL nucleation. The contribu-tion of BL nucleation toN50 andN100 was difficult to de-tect within the uncertainty of the observations, but we foundby including BL nucleation a small but significant differencewas removed at 7 and 4 of the 15 sites forN50 andN100 re-spectively.
Despite the apparent model-observation agreement ona monthly-mean basis, our analysis showed that analysis ofaggregated datasets (e.g. monthly-mean at individual sitesor multi-site means) can be misleading. The model ade-quately captures the monthly-mean size distribution and thefrequency distribution of particle concentrations, but on thehourly scale the model skill is poor. For example, the spa-tial correlation between monthly-meanNtot across all groundsites had anR2 of 0.71 in the BCOCsm model experiment,yet the hourly time seriesR2 values were very low (aver-age 0.08) and became lower still when BL nucleation wasincluded (average 0.05).
The apparent model skill at capturing aggregated datasets,but poor performance at capturing temporal variability, needsto be taken into account in model evaluations. The poortemporal correlation between model and observations willbe partly due to subgrid-scale stochastic processes such aschanges in air mass, which the model is not able to capturedue to its spatial resolution. However, there are deterministicprocesses important for particle number concentrations on anhourly scale that a global model is capable of capturing. Forexample, Spracklen et al. (2006) show good temporal agree-ment between GLOMAP and observations at the Hyytialasurface site, which were driven by nucleation events and BL
variability. To determine why the model is not as successfulfor this period would require more detailed analysis of thetemporal variability at individual sites.
It is clear that for the conditions of May 2008 the modelis unable to adequately capture the high variability observedin particle concentrations at the small end of the size distri-bution with all four BL nucleation mechanisms. Relativelygood agreement between modelled and observed condensa-tion sink at all sites suggested that the poor agreement wasnot due to the nucleation sink term, but more likely a re-sult of an underprediction of sulphuric acid, which led to anunderprediction of nucleation events during the first half ofthe IOP. The poor temporal agreement between modelled andobservedNtot precludes any attempt to identify the best nu-cleation mechanism from such a short dataset.
The temporal correlation between model and observationsincreased with particle size, and the temporal pattern ofN100(average 0.25) was in better agreement with the observationsthan ofNtot. This suggests that concentrations of particlesat sizes relevant for CCN are driven mainly by processesother than BL nucleation, which the model captures reason-ably well. The relatively good agreement between modelledand observed hourly-mean and campaign-meanN100 sug-gests that the contribution of BL nucleation needed to explainthe observedN100 is fairly small. However, the fingerprint ofnucleation is hard to detect given the S/DMPS measurementuncertainty and the uncertainties in modelling of nucleationand precursor fields and in primary carbonaceous emissions.Thus, conclusions about the role of BL nucleation still haveto be treated carefully.
There is large uncertainty associated with the prescribedsize distribution of anthropogenic carbonaceous (BC andOC) particle emissions in regional and global aerosol mod-els. The assumption of an initial size distribution for primaryparticles is necessary to account for both the size of particlesat emission and sub-grid scale aerosol processes that influ-ence the size and number concentrations of particles shortlyafter emission (Jacobson and Seinfeld, 2004; Pierce et al.,2009). However, information on the effective emission sizedistribution of carbonaceous aerosol for large-scale models islacking in mass-based emission inventories (e.g. Cooke et al.,1999; Bond et al., 2004), and thus far has only been providedby AEROCOM (Dentener et al., 2006).
Based on our review of the literature, the mode diameterrecommended by Dentener et al. (2006) for fossil fuel car-bonaceous aerosol, appears to be smaller than that which isobserved for directly emitted BC particles from traffic andurban emissions. In addition, taking into account sub-gridscale processing, the AEROCOM-recommended emissionsizes may be too small to be appropriate for large modelgrid boxes. However, when we emit BC+OC particles atsmall sizes (as recommended by AEROCOM; experimentBCOC sm), the agreement with observed particle concen-trations is generally much better than when we emit largerparticles (experiment BCOClg) that may be more realistic
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12031
for low spatial resolution global models. Our analysis hasnot been able to resolve this issue.
The agreement between experiment BCOClg and the ob-servations is generally improved when BL nucleation is in-cluded in the model. It is likely that the BCOCsm experi-ment is compensating for missing particles from BL nucle-ation by increasing the primary particle number, and thusagrees better with observations. However, it is difficult toconclude whether or not BL nucleation would make up forthe shortfall of experiment BCOClg since we do not ade-quately capture the observed nucleation events in this period.Even with high predicted nucleation-mode particle numberconcentrations with ORG1, BCOClg underpredictsN50 at13 out of 15 sites. Therefore, it is possible that the growthof nucleation mode particles may need to be increased tocapture this part of the size distribution if we assume largeBC+OC particles.
The simplified SOA formation scheme used in the model,where a fixed fraction of the oxidised products of biogenicmonoterpenes form SOA (neglecting contributions from an-thropogenic volatile or intermediate-volatile organic com-pounds), may have large implications for the growth rate andsurvival of the particles formed by homogeneous nucleation.In addition, measurements of particle number concentrationsat the surface and aloft may have been influenced by nitrateaerosol, particularly over NW Europe (Morgan et al., 2010),which the model does not account for. The role of nitrateand SOA from anthropogenic sources need to be evaluatedin future modelling studies.
Comparisons with aircraft measurements of non-volatileparticle number concentrations in the BL, suggest that themodel is missing a fairly large fraction of non-volatile parti-cles, particularly if a large emission size of primary BC+OCparticles is more appropriate for our model. Good agreementwith the observations (NMB =−3 %) was achieved only bydecreasing the carbonaceous particle emission sizes to unre-alistically small values (increasing the emitted number con-centration by∼70 % relative to BCOCsm and by more thana factor of 6 relative to BCOClg). Non-volatile residuals e.g.from mineral dust, sea spray, or BL nucleation that have notbeen included in the simulated non-volatile particle numberconcentration may partly explain the underprediction. Withthe simple representation of emitted carbonaceous particlenumber concentrations in the model we may also be missingnon-volatile residuals from anthropogenic sources.
In a future study, we aim to improve the representationof the size and number concentration of anthropogenic pri-mary particles in our model, by implementing a new Euro-pean emission inventory from EUCAARI (Denier van derGon et al., 2010). The emission inventory is based on emit-ted particle number concentration and size rather than aerosolmass. We will no longer need to assume a fixed lognormalsize distribution for primary carbonaceous emissions, thusreducing the uncertainty associated with the initial size ofBC+OC particles appropriate for a global model. The parti-
cle number emission inventory will also better represent theinhomogeneity of source size and how this influences the par-ticle number size distribution in the BL.
It is important to note there are processes in addition toBL nucleation, condensational growth and primary particleemissions that influence the particle number size distribu-tion over Europe, such as cloud processing, wet/dry depo-sition, coagulation, dynamics of semi-volatile size distribu-tions etc. These processes have not been explored in detail inthis study, but we are working towards a more complete un-certainty analysis of the GLOMAP aerosol model (Lee et al.,2011).
Acknowledgements.This research was supported by fundingfrom the EU FP6 European Integrated Project on Aerosol CloudClimate and Air Quality Interactions (EUCAARI) No. 036833-2and the NERC ADIENT project. We acknowledge the EuropeanSuper-sites for Atmospheric Aerosol Research (EUSAAR),German Ultrafine Aerosol Network (GUAN), and the DeutschesZentrum fur Luft- und Raumfahrt (DLR) for provision of data.The collection of number size distribution within GUAN wassupported by the German Federal Ministry for the Environment,Nature Conservation and Nuclear (BMU) grant F&E 370343200(Title: “Erfassung der Zahl feiner und ultrafeiner Partikel in derAußenluft”). Special thanks to Thomas Elste and George Stange(DWD, Hohenpeissenberg) for H2SO4 measurements in Melpitz.The authors would also like to thank Amy J. Braverman, SeniorStatistician at the California Institute of Technology, for guidanceon statistical time series analysis.
Edited by: V.-M. Kerminen
References
Adams,P. J. and Seinfeld, J. H.: Predicting global aerosol size dis-tributions in general circulation models, J. Geophys. Res., 107,4370, doi:10.1029/2001JD001010, 2002.
Andres, R. J. and Kasgnoc, A. D.: A time-averaged inventoryof subaerial volcanic sulfur emissions, J. Geophys. Res., 103,25251, doi:10.1029/98JD02091, 1998.
Anttila, T., Vehkamaki, H., Napari, I., and Kulmala, M.: Effect ofammonium bisulphate formation on atmospheric water-sulphuricacid-ammonia nucleation, Boreal Environ. Res., 10, 511–523,2005.
Arnold, S. R., Chipperfield, M. P., and Blitz, M. A.: A three-dimensional model study of the effect of new temperature-dependent quantum yields for acetone photolysis, J. Geophys.Res., 110, D22305, doi:10.1029/2005JD005998, 2005.
Asmi, A., Wiedensohler, A., Laj, P., Fjaeraa, A.-M., Sellegri, K.,Birmili, W., Weingartner, E., Baltensperger, U., Zdimal, V.,Zikova, N., Putaud, J.-P., Marinoni, A., Tunved, P., Hansson, H.-C., Fiebig, M., Kiveks, N., Lihavainen, H., Asmi, E., Ulevicius,V., Aalto, P. P., Swietlicki, E., Kristensson, A., Mihalopoulos,N., Kalivitis, N., Kalapov, I., Kiss, G., de Leeuw, G., Henz-ing, B., Harrison, R. M., Beddows, D., O’Dowd, C., Jennings,S. G., Flentje, H., Weinhold, K., Meinhardt, F., Ries, L., and
12032 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
Kulmala, M.: Number size distributions and seasonality of sub-micron particles in Europe 2008–2009, Atmos. Chem. Phys., 11,5505–5538, doi:10.5194/acp-11-5505-2011, 2011.
Baltensperger, U., Streit, N., Weingartner, E., Nyeki, S., Prevot, A.S. H., Van Dingenen, R., Virkkula, A., Putaud, J. P., Even, A.,Ten Brink, H., Blatter, A., Neftel, A., and Gaggeler, H. W.: Urbanand rural aerosol characterization of summer smog events duringthe PIPAPO field campaign in Milan, Italy, J. Geophys. Res.,107, 8193, doi:10.1029/2001JD001292,2002.
Bauer, S. E., Menon, S., Koch, D., Bond, T. C., and Tsigaridis, K.:A global modeling study on carbonaceous aerosol microphysi-cal characteristics and radiative effects, Atmos. Chem. Phys., 10,7439–7456, doi:10.5194/acp-10-7439-2010,2010.
Benkovitz, C. M., Trevor Scholtz, M., Pacyna, J., Tarrason, L.,Dignon, J., Voldner, E. C., Spiro, P. A., Logan, J. A., andGraedel, T. E.: Global gridded inventories of anthropogenicemissions of sulfur and nitrogen, J. Geophys. Res., 101, 29239–29253, 1996.
Berresheim, H., Elste, T., Plass-Dulmer, C., Eiseleb, F. L., and Tan-nerb, D. J.: Chemical ionization mass spectrometer for long-termmeasurements of atmospheric OH and H2SO4, Int. J. Mass Spec-trom., 202, 91–109, 2000.
Birmili, W., Weinhold, K., Nordmann, S., Weidensohler, A.,Spindler, G., Muller, K., Herrmann, H., Gnauk, T., Pitz, M.,Cyrya, J., Flentje, H., Nickel, C., Kuhlbusch, T. A. J.,Loschau, G., Haase, D., Meinhardt, F., Schwerin, A., Ries, L.,and Wirtz, K.: Atmospheric aerosol measurements in the Ger-man Ultrafine Aerosol Network (GUAN) part 1: soot and particlenumber size distributions, Gefahrst. Reinhalt. L., 69, 137–145,2009.
Birmili, W., Heinke, K., Pitz, M., Matschullat, J., Wiedensohler,A., Cyrys, J., Wichmann, H.-E., and Peters, A.: Particle num-ber size distributions in urban air before and after volatilisation,Atmos. Chem. Phys., 10, 4643–4660, doi:10.5194/acp-10-4643-2010,2010.
Bond,T. C., Streets, D. G., Yarber, K. F., Nelson, S. M., Woo, J. H.,and Klimont, Z.: A technology-based global inventory of blackand organic carbon emissions from combustion, J. Geophys.Res., 109, D14203, doi:10.1029/2003JD003697,2004.
Bond,T. C., Habib, G., and Bergstrom, R. W.: Limitations in the en-hancement of visible light absorption due to mixing state, J. Geo-phys. Res., 111, D20211, doi:10.1029/2006JD007315,2006.
Boy, M., Kazil, J., Lovejoy, E. R., Guenther, A., and Kulmala, M.:Relevance of ion-induced nucleation of sulfuric acid and water inthe lower troposphere over the boreal forest at northern latitudes,Atmos. Res., 90, 151–158, 2008.
Burtscher, H., Baltensperger, U., Bukowiecki, N., Cohn, P.,Huglin, C., Mohr, M., Matter, U., Nyeki, S., Schmatloch, V.,Streit, N., and Weingartner, E.: Separation of volatile and non-volatile aerosol fractions by thermodesorption: instrumental de-velopment and applications, J. Aerosol Sci., 32, 427–442, 2001.
Casati, R., Scheer, V., Vogt, R., and Benter, T.: Measurements ofnucleation and soot mode particle emission from a diesel pas-senger car in real world and laboratory in situ dilution, Atmos.Environ., 41, 2125–2135, 2007.
Chipperfield, M. P.: New version of the TOMCAT/SLIMCAT off-line chemical transport model: Intercomparison of stratospherictracer experiments, Q. J. Roy. Meteor. Soc., 132, 1179–1203,2006.
Cofala, J., Amann, M., Klimont, Z., and Schopp, W.: Scenariosof World Anthropogenic Emissions of SO2, NOx, and CO upto 2030, in: Internal report of the Transboundary Air PollutionProgramme, p. 17, International Institute for Applied SystemsAnalysis, Laxenburg, Austria, 2005.
Cooke, W. F., Liousse, C., Cachier, H., and Feichter, J.: Con-struction of a 1×1 fossil fuel emission data set for carbona-ceous aerosol and implementation and radiative impact in theECHAM4 model, J. Geophys. Res., 104, 22137–22162, 1999.
Costabile,F., Birmili, W., Klose, S., Tuch, T., Wehner, B., Wieden-sohler, A., Franck, U., Konig, K., and Sonntag, A.: Spatio-temporal variability and principal components of the particlenumber size distribution in an urban atmosphere, Atmos. Chem.Phys., 9, 3163–3195, doi:10.5194/acp-9-3163-2009, 2009.
Deniervan der Gon, H., Visschedijk, A., Johansson, C., Ntziachris-tos, L., and Harrison, R.: Particle number inventory, Abstract1E1, International Aerosol Conference 2010, Helsinki, Finland,29 August–3 September, 2010.
Dentener, F., Kinne, S., Bond, T., Boucher, O., Cofala, J., Generoso,S., Ginoux, P., Gong, S., Hoelzemann, J. J., Ito, A., Marelli, L.,Penner, J. E., Putaud, J.-P., Textor, C., Schulz, M., van der Werf,G. R., and Wilson, J.: Emissions of primary aerosol and pre-cursor gases in the years 2000 and 1750 prescribed data-sets forAeroCom, Atmos. Chem. Phys., 6, 4321–4344, doi:10.5194/acp-6-4321-2006, 2006.
Durbin, J. and Watson, G. S.: Testing for serial correlation in leastsquares regression: I, Biometrika, 37, 409–428, 1950.
Ehn,M., Petaja, T., Birmili, W., Junninen, H., Aalto, P., and Kul-mala, M.: Non-volatile residuals of newly formed atmosphericparticles in the boreal forest, Atmos. Chem. Phys., 7, 677–684,doi:10.5194/acp-7-677-2007, 2007.
Elleman,R. A. and Covert, D. S.: Aerosol size distribution model-ing with the Community Multiscale Air Quality modeling systemin the Pacific Northwest: 2. Parameterizations for ternary nu-cleation and nucleation mode processes, J. Geophys. Res., 114,D11207, doi:10.1029/2009JD012187, 2009.
Engler, C., Rose, D., Wehner, B., Wiedensohler, A., Bruggemann,E., Gnauk, T., Spindler, G., Tuch, T., and Birmili, W.: Size distri-butions of non-volatile particle residuals (Dp < 800 nm) at a ru-ral site in Germany and relation to air mass origin, Atmos. Chem.Phys., 7, 5785–5802, doi:10.5194/acp-7-5785-2007, 2007.
Fuchs,N. A. and Sutugin, A. G.: High-dispersed aerosols, in: Top-ics in Current Aerosol Research, edited by: Hidy, G. M. andBrock, J. R., vol. 2, Pergamon Press, Oxford, 1–60, 1971.
Geller, M. D., Sardar, S. B., Phuleria, H., Philip, M., and Sioutas, C.:Measurements of particle number and mass concentrations andsize distributions in a tunnel environment, Environ. Sci. Technol.,39, 8653–8663, 2005.
Gilardoni,S., Vignati, E., and Wilson, J.: Using measurements forevaluation of black carbon modeling, Atmos. Chem. Phys., 11,439–455, doi:10.5194/acp-11-439-2011, 2011.
Gong, S. L., Barrie, L. A., Blanchet, J. P., Von Salzen, K.,Lohmann, U., Lesins, G., Spacek, L., Zhang, L. M., Girard, E.,Lin, H., Leaitch, R., Leighton, H., Chylek, P., and Huang, P.:Canadian Aerosol Module: a size-segregated simulation of at-mospheric aerosol processes for climate and air quality mod-els 1. Module development, J. Geophys. Res., 108, 4007,doi:10.1029/2001JD002002, 2003.
Guenther, A., Hewitt, C. N., Erickson, D., Fall, R., Geron, C.,
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12033
Graedel, T., Harley, P., Klinger, L., Lerdau, M., McKay, W. A.,Pierce, T., Scholes, B., Steinbrecher, R., Tallamraju, R., Tay-lor, J., and Zimmerman, P.: A global model of natural volatileorganic compound emissions, J. Geophys. Res., 100, 8873–8892,1995.
Hamburger, T., McMeeking, G., Minikin, A., Birmili, W.,Dall’Osto, M., O’Dowd, C., Flentje, H., Henzing, B., Junninen,H., Kristensson, A., de Leeuw, G., Stohl, A., Burkhart, J. F.,Coe, H., Krejci, R., and Petzold, A.: Overview of the synop-tic and pollution situation over Europe during the EUCAARI-LONGREX field campaign, Atmos. Chem. Phys., 11, 1065–1082, doi:10.5194/acp-11-1065-2011,2011.
Harris, S. J. and Maricq, M. M.: Signature size distributions fordiesel and gasoline engine exhaust particulate matter, J. AerosolSci., 32, 749–764, 2001.
Henne, S., Brunner, D., Folini, D., Solberg, S., Klausen, J., andBuchmann, B.: Assessment of parameters describing representa-tiveness of air quality in-situ measurement sites, Atmos. Chem.Phys., 10, 3561–3581, doi:10.5194/acp-10-3561-2010,2010.
Henze,D. K., Seinfeld, J. H., Ng, N. L., Kroll, J. H., Fu, T.-M.,Jacob, D. J., and Heald, C. L.: Global modeling of secondaryorganic aerosol formation from aromatic hydrocarbons: high-vs. low-yield pathways, Atmos. Chem. Phys., 8, 2405–2420,doi:10.5194/acp-8-2405-2008,2008.
Holtslag, A. A. M. and Boville, B. A.: Local versus nonlocalboundary-layer diffusion in a global climate model, J. Climate,6, 1825–1842, 1993.
Jacobson, M. Z. and Seinfeld, J. H.: Evolution of nanoparticle sizeand mixing state near the point of emission, Atmos. Environ., 38,1839–1850, 2004.
Jennings, S. G. and O’Dowd, C. D.: Volatility of aerosol at MaceHead, on the west coast of Ireland, J. Geophys. Res., 95, 13937,doi:10.1029/JD095iD09p13937,1990.
Abdul-Khalek,I. S., Kittelson, D. B., and Brear, F.: NanoparticleGrowth During Dilution and Cooling of Diesel Exhaust: Exper-imental Investigation and Theoretical Assessment, SAE Techni-cal Paper 2000-01-1515, 2000.
Kerminen, V.-M. and Kulmala, M.: Analytical formulae connectingthe “real” and the “apparent” nucleation rate and the nuclei num-ber concentration for atmospheric nucleation events, J. AerosolSci., 33, 609–622, doi:10.1016/S0021-8502(01)00194-X,2002.
Kerminen, V.-M., Lihavainen, H., Komppula, M., Viisanen, Y., andKulmala, M.: Direct observational evidence linking atmosphericaerosol formation and cloud droplet activation, Geophys. Res.Lett., 32, L14803, doi:10.1029/2005GL023130,2005.
Kerminen, V.-M., Petaja, T., Manninen, H. E., Paasonen, P., Niem-inen, T., Sipila, M., Junninen, H., Ehn, M., Gagne, S., Laakso,L., Riipinen, I., Vehkamaki, H., Kurten, T., Ortega, I. K., DalMaso, M., Brus, D., Hyvarinen, A., Lihavainen, H., Leppa, J.,Lehtinen, K. E. J., Mirme, A., Mirme, S., Horrak, U., Berndt, T.,Stratmann, F., Birmili, W., Wiedensohler, A., Metzger, A., Dom-men, J., Baltensperger, U., Kiendler-Scharr, A., Mentel, T. F.,Wildt, J., Winkler, P. M., Wagner, P. E., Petzold, A., Minikin, A.,Plass-Dulmer, C., Poschl, U., Laaksonen, A., and Kulmala, M.:Atmospheric nucleation: highlights of the EUCAARI projectand future directions, Atmos. Chem. Phys., 10, 10829–10848,doi:10.5194/acp-10-10829-2010,2010.
Kettle, A. J. and Andreae, M. O.: Flux of dimethylsulfide from theoceans: a comparison of updated data sets and flux models, J.
Geophys. Res., 105, 26793–26808, 2000.Kittelson, D. B.: Engines and nanoparticles: a review, J. Aerosol
Sci., 29, 575–588, 1998.Kittelson, D. B., Johnson, J., Watts, W., Wei, Q., Drayton, M.,
Paulsen, D., and Bukowiecki, N.: Diesel aerosol sampling in theatmosphere, Society of Automotive Engineers, Technical PaperSeries, No. 2000-01-2212,doi:10.4271/2000-01-2212, 2000.
Kittelson,D. B., Watts, W. F., and Johnson, J. P.: On-road and lab-oratory evaluation of combustion aerosols – Part1: Summary ofdiesel engine results, J. Aerosol Sci., 37, 913–930, 2006.
Kuang, C., McMurry, P. H., and McCormick, A. V., Determi-nation of cloud condensation nuclei production from measurednew particle formation events, Geophys. Res. Lett., 36, L09822,doi:10.1029/2009GL037584, 2009.
Kulmala, M. and Laaksonen, A.: Binary nucleation of water–sulfuric acid system: comparison of classical theories with differ-ent H2SO4 saturation vapor pressures, J. Chem. Phys., 93, 696–701, 1990.
Kulmala, M., Laaksonen, A., and Pirjola, L.: Parameterizations forsulfuric acid/water nucleation rates, J. Geophys. Res., 103, 8301–8307, 1998.
Kulmala, M., Pirjola, L., and Makela, J. M.: Stable sulphate clustersas a source of new atmospheric particles, Nature, 404, 66–69,2000.
Kulmala, M., Dal Maso, M., Makela, J. M., Pirjola, L., Vakeva, M.,Aalto, P., Miikkulainen, P., Hameri, K., and O’Dowd, C. D.:On the formation, growth and composition of nucleationmode particles, Tellus B, 53, 479–490,doi:10.1034/j.1600-0889.2001.530411.x, 2001.
Kulmala, M., Vehkamaki, H., Petaja, T., Dal Maso, M., Lauri, A.,Kerminen, V.-M., Birmili, W., and McMurry, P. H.: Formationand growth rates of ultrafine atmospheric particles: a review ofobservations, J. Aerosol Sci., 35, 143–176, 2004.
Kulmala, M., Lehtinen, K. E. J., and Laaksonen, A.: Cluster activa-tion theory as an explanation of the linear dependence betweenformation rate of 3 nm particles and sulphuric acid concentration,Atmos. Chem. Phys., 6, 787–793, doi:10.5194/acp-6-787-2006,2006.
Kulmala, M., Riipinen, I., Sipila, M., Manninen, H. E., Petaja, T.,Junninen, H., Maso, M. D., Mordas, G., Mirme, A., Vana, M.,Hirsikko, A., Laakso, L., Harrison, R. M., Hanson, I., Le-ung, C., Lehtinen, K. E. J., and Kerminen, V.-M.: Toward directmeasurement of atmospheric nucleation, Science, 318, 89–92,doi:10.1126/science.1144124, 2007.
Kulmala, M., Asmi, A., Lappalainen, H. K., Carslaw, K. S., Poschl,U., Baltensperger, U., Hov, Ø., Brenquier, J.-L., Pandis, S.N., Facchini, M. C., Hansson, H.-C., Wiedensohler, A., andO’Dowd, C. D.: Introduction: European Integrated Project onAerosol Cloud Climate and Air Quality interactions (EUCAARI)integrating aerosol research from nano to global scales, At-mos. Chem. Phys., 9, 2825–2841, doi:10.5194/acp-9-2825-2009,2009.
Laakso,L., Makela, J. M., Pirjola, L., and Kulmala, M.: Modelstudies on ion-induced nucleation in the atmosphere, J. Geophys.Res., 107, 4427, doi:10.1029/2002JD002140, 2002.
Laakso,L., Gagne, S., Petaja, T., Hirsikko, A., Aalto, P. P., Kulmala,M., and Kerminen, V.-M.: Detecting charging state of ultra-fineparticles: instrumental development and ambient measurements,Atmos. Chem. Phys., 7, 1333–1345, doi:10.5194/acp-7-1333-
12034 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
2007, 2007.Laaksonen,A., Hamed, A., Joutsensaari, J., Hiltunen, L., Cav-
alli, F., Junkermann, W., Asmi, A., Fuzzi, S., and Facchini, M. C.:Cloud condensation nucleus production from nucleation eventsat a highly polluted region, Geophys. Res. Lett., 32, L06812,doi:10.1029/2004GL022092,2005.
Lee,L. A., Carslaw, K. S., Pringle, K., Mann, G. W., and Spracklen,D. V.: Emulation of a complex global aerosol model to quan-tify sensitivity to uncertain parameters, Atmos. Chem. Phys.Discuss., 11, 20433–20485, doi:10.5194/acpd-11-20433-2011,2011.
Lihavainen, H., Kerminen, V.-M., Komppula, M., Hatakka, J., Aal-tonen, V., Kulmala, M., and Viisanen, Y.: Production of “poten-tial” cloud condensation nuclei associated with atmospheric new-particle formation in northern Finland, J. Geophys. Res., 108,4782, doi:10.1029/2003JD003887,2003.
Liu, P. S. K., Leaitch, W. R., Strapp, J. W., and Wasey, M. A.:Response of particle measuring systems airborne ASASP andPCASP to NaCl and latex particles, Aerosol Sci. Tech., 16, 83–95, 1992.
Lohmann, U. and Feichter, J.: Global indirect aerosol effects: a re-view, Atmos. Chem. Phys., 5, 715–737, doi:10.5194/acp-5-715-2005,2005.
Lucas,D. D. and Akimoto, H.: Evaluating aerosol nucleation pa-rameterizations in a global atmospheric model, Geophys. Res.Lett., 33, L10808, doi:10.1029/2006GL025672,2006.
McMurry, P. and Friedlander, S.: New particle formation in thepresence of an aerosol, Atmos. Environ., 13, 1635–1651, 1979.
Merikanto, J., Napari, I., Vehkamaki, H., Anttila, T., and Kul-mala, M.: New parameterization of sulfuric acid-ammonia-waterternary nucleation rates at tropospheric conditions, J. Geophys.Res., 112, D15207, doi:10.1029/2006JD007977,2007.
Merikanto, J., Spracklen, D. V., Mann, G. W., Pickering, S. J.,and Carslaw, K. S.: Impact of nucleation on global CCN, At-mos. Chem. Phys., 9, 8601–8616, doi:10.5194/acp-9-8601-2009,2009.
Metzger, A., Verheggen, B., Dommen, J., Duplissy, J., Pre-vot, A. S. H., Weingartner, E., Riipinen, I., Kulmala, M.,Spracklen, D. V., Carslaw, K. S., and Baltensperger, U.: Evi-dence for the role of organics in aerosol particle formation un-der atmospheric conditions, P. Natl. Acad. Sci. USA, 107, 6646–6651, doi:10.1073/pnas.0911330107,2010.
Modgil, M. S., Kumar, S., Tripathi, S. N., and Lovejoy, E. R.:A parameterization of ion-induced nucleation of sulphuric acidand water for atmospheric conditions, J. Geophys. Res., 110,D19205, doi:10.1029/2004JD005475,2005.
Morgan, W. T., Allan, J. D., Bower, K. N., Esselborn, M., Har-ris, B., Henzing, J. S., Highwood, E. J., Kiendler-Scharr, A.,McMeeking, G. R., Mensah, A. A., Northway, M. J., Osborne,S., Williams, P. I., Krejci, R., and Coe, H.: Enhancement of theaerosol direct radiative effect by semi-volatile aerosol compo-nents: airborne measurements in North-Western Europe, Atmos.Chem. Phys., 10, 8151–8171, doi:10.5194/acp-10-8151-2010,2010.
Nightingale, P. D., Liss, P. S., and Schlosser, P.: Measurements ofair-sea gas transfer during an open ocean algal bloom, Geophys.Res. Lett., 27, 2117–2120, 2000.
O’Dowd, C. D. and Smith, M. H.: Physicochemical properties ofaerosols over the Northeast Atlantic: evidence for wind-speed-
related submicron sea-salt aerosol production, J. Geophys. Res.,98, 1137–1149, 1993.
Paasonen, P., Sihto, S.-L., Nieminen, T., Vuollekoski, H., Riip-inen, I., Plaß-Dulmer, C., Berresheim, H., Birmili, W., and Kul-mala, M.: Connection between new particle formation and sul-phuric acid at Hohenpeissenberg (Germany) including the influ-ence of organic compounds, Boreal Environ. Res., 14, 616–629,2009.
Paasonen, P., Nieminen, T., Asmi, E., Manninen, H. E., Petaja, T.,Plass-Dulmer, C., Flentje, H., Birmili, W., Wiedensohler, A.,Horrak, U., Metzger, A., Hamed, A., Laaksonen, A., Facchini,M. C., Kerminen, V.-M., and Kulmala, M.: On the roles of sul-phuric acid and low-volatility organic vapours in the initial stepsof atmospheric new particle formation, Atmos. Chem. Phys., 10,11223–11242, doi:10.5194/acp-10-11223-2010, 2010.
Pierce,J. R. and Adams, P. J.: Efficiency of cloud condensationnuclei formation from ultrafine particles, Atmos. Chem. Phys.,7, 1367–1379, doi:10.5194/acp-7-1367-2007, 2007.
Pierce,J. R. and Adams, P. J.: Uncertainty in global CCN concen-trations from uncertain aerosol nucleation and primary emissionrates, Atmos. Chem. Phys., 9, 1339–1356, doi:10.5194/acp-9-1339-2009, 2009.
Pierce,J. R., Theodoritsi, G., Adams, P. J., and Pandis, S. N.: Pa-rameterization of the effect of sub-grid scale aerosol dynamicson aerosol number emission rates, J. Aerosol Sci., 40, 385–393,2009.
Putaud,J. P., Raes, F., Van Dingenen, R., Bruggemann, E., Fac-chini, M. C., Decesari, S., Fuzzi, S., Gehrig, R., Huglin, C.,Laj, P., Lorbeer, G., Maenhaut, W., Mihalopoulos, N., Muller, K.,Querol, X., Rodriguez, S., Schneider, J., Spindler, G., ten-Brink, H., Torseth, K., and Wiedensohler, A.: A Europeanaerosol phenomenology – 2: chemical characteristics of partic-ulate matter at kerbside, urban, rural and background sites in Eu-rope, Atmos. Environ., 38, 2579–2595, 2004.
Ramanathan,V., Crutzen, P. J., Kiehl, J. T., and Rosenfeld, D.:Aerosols, climate, and the hydrological cycle, Science, 294,2119–2124, doi:10.1126/science.1064034, 2001.
Riipinen, I., Sihto, S.-L., Kulmala, M., Arnold, F., Dal Maso, M.,Birmili, W., Saarnio, K., Teinil, K., Kerminen, V.-M., Laakso-nen, A., and Lehtinen, K. E. J.: Connections between atmo-spheric sulphuric acid and new particle formation during QUESTIII–IV campaigns in Heidelberg and Hyytiala, Atmos. Chem.Phys., 7, 1899–1914, doi:10.5194/acp-7-1899-2007, 2007.
Roldin, P., Swietlicki, E., Massling, A., Kristensson, A., Londahl,J., Eriksson, A., Pagels, J., and Gustafsson, S.: Aerosol ageing inan urban plume - implication for climate, Atmos. Chem. Phys.,11, 5897–5915, doi:10.5194/acp-11-5897-2011, 2011.
Robinson,A. L., Donahue, N. M., Shrivastava, M. K., Weitkamp,E. A., Sage, A. M., Grieshop, A. P., Lane, T. E., Pierce, J. R.,and Pandis, S. N.: Rethinking organic aerosols: Semivolatileemissions and photochemical ageing, Science, 315, 1259–1262,2007.
Rose,D., Wehner, B., Ketzel, M., Engler, C., Voigtlander, J., Tuch,T., and Wiedensohler, A.: Atmospheric number size distribu-tions of soot particles and estimation of emission factors, At-mos. Chem. Phys., 6, 1021–1031, doi:10.5194/acp-6-1021-2006,2006.
Sgro,L. A., Borghese, A., Speranza, L., Barone, A. C., Minutolo,P., Bruno, A., Danna, A., and DAlessio, A.: Measurements of
C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations 12035
nanoparticles of organic carbon and soot in flames and vehicleexhausts, Environ. Sci. Technol., 42, 859–863, 2008.
Sihto,S.-L., Kulmala, M., Kerminen, V.-M., Dal Maso, M., Petaja,T., Riipinen, I., Korhonen, H., Arnold, F., Janson, R., Boy, M.,Laaksonen, A., and Lehtinen, K. E. J.: Atmospheric sulphuricacid and aerosol formation: implications from atmospheric mea-surements for nucleation and early growth mechanisms, At-mos. Chem. Phys., 6, 4079–4091, doi:10.5194/acp-6-4079-2006,2006.
Sipila, M., Berndt, T., Petaja, T., Brus, D., Vanhanen, J., Strat-mann, F., Patokoski, J., Mauldin, R. L., Hyvarinen, A.-P., Lihavainen, H., and Kulmala, M.: The role of sulfu-ric acid in atmospheric nucleation, Science, 327, 1243–1246,doi:10.1126/science.1180315,2010.
Spracklen,D. V., Pringle, K. J., Carslaw, K. S., Chipperfield,M. P., and Mann, G. W.: A global off-line model of size-resolved aerosol microphysics: I. Model development and pre-diction of aerosol properties, Atmos. Chem. Phys., 5, 2227–2252, doi:10.5194/acp-5-2227-2005,2005a.
Spracklen,D. V., Pringle, K. J., Carslaw, K. S., Chipperfield, M.P., and Mann, G. W.: A global off-line model of size-resolvedaerosol microphysics: II. Identification of key uncertainties, At-mos. Chem. Phys., 5, 3233–3250, doi:10.5194/acp-5-3233-2005,2005b.
Spracklen,D. V., Carslaw, K. S., Kulmala, M., Kerminen, V.-M.,Mann, G. W., and Sihto, S.-L.: The contribution of boundarylayer nucleation events to total particle concentrations on re-gional and global scales, Atmos. Chem. Phys., 6, 5631–5648,doi:10.5194/acp-6-5631-2006,2006.
Spracklen,D. V., Carslaw, K. S., Kulmala, M., Kerminen, V.-M.,Sihto, S.-L., Riipinen, I., Merikanto, J., Mann, G. W., Chipper-field, M. P., Wiedensohler, A., Birmili, W., and Lihavainen, H.:Contribution of particle formation to global cloud condensa-tion nuclei concentrations, Geophys. Res. Lett., 35, L06808,doi:10.1029/2007GL033038,2008.
Spracklen,D. V., Carslaw, K. S., Merikanto, J., Mann, G. W., Red-dington, C. L., Pickering, S., Ogren, J. A., Andrews, E., Bal-tensperger, U., Weingartner, E., Boy, M., Kulmala, M., Laakso,L., Lihavainen, H., Kivekas, N., Komppula, M., Mihalopoulos,N., Kouvarakis, G., Jennings, S. G., O’Dowd, C., Birmili, W.,Wiedensohler, A., Weller, R., Gras, J., Laj, P., Sellegri, K., Bonn,B., Krejci, R., Laaksonen, A., Hamed, A., Minikin, A., Harri-son, R. M., Talbot, R., and Sun, J.: Explaining global surfaceaerosol number concentrations in terms of primary emissionsand particle formation, Atmos. Chem. Phys., 10, 4775–4793,doi:10.5194/acp-10-4775-2010,2010.
Spracklen,D. V., Carslaw, K. S., Poschl, U., Rap, A., and Forster,P. M.: Global cloud condensation nuclei influenced by carbona-ceous combustion aerosol, Atmos. Chem. Phys., 11, 9067–9087,doi:10.5194/acp-11-9067-2011,2011.
Stier, P., Feichter, J., Kinne, S., Kloster, S., Vignati, E., Wilson,J., Ganzeveld, L., Tegen, I., Werner, M., Balkanski, Y., Schulz,M., Boucher, O., Minikin, A., and Petzold, A.: The aerosol-climate model ECHAM5-HAM, Atmos. Chem. Phys., 5, 1125–1156, doi:10.5194/acp-5-1125-2005,2005.
Textor, C., Schulz, M., Guibert, S., Kinne, S., Balkanski, Y., Bauer,S., Berntsen, T., Berglen, T., Boucher, O., Chin, M., Dentener,F., Diehl, T., Easter, R., Feichter, H., Fillmore, D., Ghan, S., Gi-noux, P., Gong, S., Grini, A., Hendricks, J., Horowitz, L., Huang,
P., Isaksen, I., Iversen, I., Kloster, S., Koch, D., Kirkevag, A.,Kristjansson, J. E., Krol, M., Lauer, A., Lamarque, J. F., Liu,X., Montanaro, V., Myhre, G., Penner, J., Pitari, G., Reddy, S.,Seland, Ø., Stier, P., Takemura, T., and Tie, X.: Analysis andquantification of the diversities of aerosol life cycles within Ae-roCom, Atmos. Chem. Phys., 6, 1777–1813, doi:10.5194/acp-6-1777-2006, 2006.
Van Dingenen, R., Raes, F., Putaud, J.-P., Baltensperger, U., Char-ron, A., Facchini, M. C., Decesari, S., Fuzzi, S., Gehrig, R.,Hansson, H.-C., Harrison, R. M., Huglin, C., Jones, A. M.,Laj, P., Lorbeer, G., Maenhaut, W., Palmgren, F., Querol, X., Ro-driguez, S., Schneider, J., ten Brink, H., Tunved, P., Tørseth, K.,Wehner, B., Weingartner, E., Wiedensohler, A., and Wahlin, P.:A European aerosol phenomenology – 1: physical characteristicsof particulate matter at kerbside, urban, rural and backgroundsites in Europe, Atmos. Environ., 38, 2561–2577, 2004.
Vehkamaki, H., Kulmala, M., Napari, I., Lehtinen, K. E. J., Timm-reck, C., Noppel, M., and Laaksonen, A.: An improved pa-rameterization for sulfuric acid-water nucleation rates for tro-pospheric and stratospheric conditions, J. Geophys. Res., 107,4622, doi:10.1029/2002JD002184, 2002.
Venzac, H., Sellegri, K., Villani, P., Picard, D., and Laj, P.: Seasonalvariation of aerosol size distributions in the free troposphere andresidual layer at the puy de Dme station, France, Atmos. Chem.Phys., 9, 1465–1478, doi:10.5194/acp-9-1465-2009, 2009.
Wang, S. C. and Flagan, R. C.: Scanning electrical mobility spec-trometer, Aerosol Sci. Tech., 13, 230–240, 1990.
Wang, M. and Penner, J. E.: Aerosol indirect forcing in a globalmodel with particle nucleation, Atmos. Chem. Phys., 9, 239–260,doi:10.5194/acp-9-239-2009, 2009.
Weber, R. J., Marti, J. J., McMurry, P. H., Eisele, F. L., Tanner, D. J.,and Jefferson, A.: Measured atmospheric new particle forma-tion rates: implications for nucleation mechanisms, Chem. Eng.Commun., 151, 53–64, 1996.
Wehner, B., Birmili, W., Gnauk, T., and Wiedensohler, A.: Particlenumber size distributions in a street canyon and their transforma-tion into the urban-air background: measurements and a simplemodel study, Atmos. Environ., 36, 2215–2223, 2002.
Wehner, B., Philippin, S., Wiedensohler, A., Scheer, V., Vogt, R.:Variability of non-volatile fractions of atmospheric aerosol par-ticles with traffic influence, Atmos. Environ., 38, 6081–6090,2004.
Wehner, B., Petaja, T., Boy, M., Engler, C., Birmili, W., Tuch, T.,Wiedensohler, A., and Kulmala, M.: The contribution of sulfu-ric acid and non-volatile compounds on the growth of freshlyformed atmospheric aerosols, Geophys. Res. Lett., 32, L17810,doi:10.1029/2005GL023827, 2005.
Wehner, B., Uhrner, U., von Lowis, S., Zallinger, M., and Wieden-sohler, A.: Aerosol number size distributions within the exhaustplume of a diesel and a gasoline passenger car under on-road con-ditions and determination of emission factors, Atmos. Environ.,43, 1235–1245, 2009.
Weimer, S., Mohr, C., Richter, R., Keller, J., Mohr, M., Prevot,A.S.H., and Baltensperger, U.: Mobile measurements of aerosolnumber and volume size distributions in an Alpine valley: Influ-ence of traffic versus wood burning, Atmos. Environ., 43, 624–630, 2009.
Weingartner, E., Nyeki, S., and Baltensperger, U.: Seasonal anddiurnal variation of aerosol size distributions (10<D<750 nm) at
12036 C. L. Reddington et al.: Primary vs. secondary contributions to PN concentrations
a high-alpine site (Jungfraujoch 3580 m a.s.l.), J. Geophys. Res.,104, 26809, doi:10.1029/1999JD900170, 1999.
Wiedensohler, A., Birmili, W., Nowak, A., Sonntag, A., Weinhold,K., Merkel, M., Wehner, B., Tuch, T., Pfeifer, S., Fiebig, M.,Fjaraa, A. M., Asmi, E., Sellegri, K., Depuy, R., Venzac, H., Vil-lani, P., Laj, P., Aalto, P., Ogren, J. A., Swietlicki, E., Roldin,P., Williams, P., Quincey, P., Hglin, C., Fierz-Schmidhauser, R.,Gysel, M., Weingartner, E., Riccobono, F., Santos, S., Gruning,C., Faloon, K., Beddows, D., Harrison, R. M., Monahan, C., Jen-nings, S. G., O’Dowd, C. D., Marinoni, A., Horn, H.-G., Keck,L., Jiang, J., Scheckman, J., McMurry, P. H., Deng, Z., Zhao, C.S., Moerman, M., Henzing, B., and de Leeuw, G.: Particle mo-bility size spectrometers: harmonization of technical standardsand data structure to facilitate high quality long-term observa-tions of atmospheric particle number size distributions, Atmos.Meas. Tech. Discuss., 3, 5521–5587, doi:10.5194/amtd-3-5521-2010,2010.
van der Werf, G. R., Randerson, J. T., Collatz, G. J., and Giglio, L.:Carbon emissions from fires in tropical and subtropical ecosys-tems, Glob. Change Biol., 9, 547–562, 2003.
Wilks, D. S.: Resampling hypothesis tests for autocorrelated fields,J. Climate, 10, 66–82, 1997.
Yu, F. and Luo, G.: Simulation of particle size distribution witha global aerosol model: contribution of nucleation to aerosoland CCN number concentrations, Atmos. Chem. Phys., 9, 7691–7710, doi:10.5194/acp-9-7691-2009,2009.
Yu, F., Luo, G., Bates, T. S., Anderson, B., Clarke, A., Ka-pustin, V., Yantosca, R. M., Wang, Y., and Wu, S.: Spa-tial distributions of particle number concentrations in theglobal troposphere: simulations, observations, and implicationsfor nucleation mechanisms, J. Geophys. Res., 115, D17205,doi:10.1029/2009JD013473, 2010.
Zhang,K. M. and Wexler, A. S.: Evolution of particle number dis-tribution near roadways Part I: analysis of aerosol dynamics andits implications for engine emission measurement. Atmos. Envi-ron., 38, 6643–6653, 2004.
Zhang,K. M., Wexler, A. S., Zhu, Y. F., Hinds, W. C., and Sioutas,C.: Evolution of particle number distributions near roadwaysPart II: the road-to-ambient process, Atmos. Environ., 38, 6655–6665, 2004.
Zhang,K. M., Wexler, A. S., Niemeier, D. A., Yi, F. Z., Hinds,W. C., and Sioutas, C.: Evolution of particle number distributionnear roadways. Part III: Traffic analysis and on-road size resolvedparticulate emission factors, Atmos. Environ., 39, 4155–4166,2005.
Zhu, Y., Hinds, W.C., Kim, S., Shen, S., and Sioutas, C.: Study ofultrafine particles near a major highway with heavy-duty dieseltraffic, Atmos. Environ. 36, 4323–4335, 2002.
